Ln. ..
1%.-.. 2 mm...

1f;

 

L. 4...! 4 w
.311 .5... r.

J

4 A ‘ u‘ E .3
21:» .29..
.VFwtza . . . mu”. ﬂu“ a\ a
. 3
:{VI ..
4:”... .t...
t .N I‘ .

 

 

 

 

 

pint...
r
a. .1

33d!- .
$1.559?

. 2 g? (3.. ’12.")...
.‘Qalﬂtt’;1l.iz..
.glggzyﬂéi. :15:
“ﬂun‘hufu. ! . .-

1 .
at}?

. .9... .

.. .tl. o.
1.:
.14

In . 315!

(“Vt . ‘l‘rK—t.
v ..l.$>ha.uurlnvnuacaivl .
1.2.0.... .5. luqummlwwwi
1|..hnﬂhhvs . ,
35

2?...1.

t.-. .5".
L‘trttrnlr‘
Ell-iii». 2
.111 3.... a #034.
14: :1... I. . J. .
_ . (IA-Sf UNI}
£01... In, WM”! .11.
9%.??? Allartrnn”.
15.7

 

 

., , .. . 1 Z , $1.... unﬁxnrmmwm.......é
, gs... , . .5.%%. a...

v 1.1..
1 i. .. . .

 

This is to certify that the
thesis entitled

WELFARE EFFECTS OF MAIZE PRICING POLICY ON
RURAL HOUSEHOLDS IN KENYA

presented by

ELLIOT WAMBOKA MGHENYI

has been accepted towards fulﬁllment
of the requirements for the

MS degree in Agicultural Economics

 

 

Major Professor’s Signhtdre
'7
AW ( 4.! 10-73 6

Date

 

MSU is an Afﬁrmative Action/Equal Opportunity Institution

 

LIBRARY
Michioan State

 

 

v v -v r-r r-v v-u-r-e-rz-w ‘r-weewz-

' V ‘. O

“' "‘—V. '

PLACE IN RETURN BOX to remove this checkout from your record.
TO AVOID FINES return on or before date due.
MAY BE RECALLED with earlier due date if requested.

 

DATE DUE DATE DUE DATE DUE

 

 

 

 

 

 

 

 

 

 

 

 

 

 

2/05 p:/ClRC/DateDue.indd-p.1

 

WELFARE EFFECTS OF MAIZE PRICING POLICY ON RURAL
HOUSEHOLDS IN KENYA

By

Elliot Wamboka Mghenyi

A MASTERS THESIS

Submitted to
Michigan State University
in partial fulﬁllment of the requirements
for the degree of

MASTER OF SCIENCE

Department of Agricultural Economics

2006

ABSTRACT

WELFARE EFFECTS OF MAIZE PRICING POLICY ON RURAL
HOUSEHOLDS IN KENYA
By
Elliot Wamboka Mghenyi
Welfare effects of food pricing are not straightforward when households are both
producers and consumers of a main staple. Using 2000 and 2004 panel data, I examine
the effects of the twin policy of maize price supports and import tariffs on poverty and
the distribution of income in rural Kenya. In chapter 3, I discuss the conceptual
framework for measuring changes in income due to the effects of the policy, leading to a
second order approximation of equilibrium income changes. The framework forms the
basis for poverty orderings between two income distributions; counterfactual incomes
and incomes with the effects of price controls. The results indicate that the policy
exacerbates poverty in all regions considered except the high potential maize zone. In the
Western transitional zone, the number of the poor may not have increased but their
income shortfalls are increased. In chapter 4, I consider the effects of the policy on the
distribution of income. First, I specify a partly linear semi-parametric model to control
for the inﬂuence of household characteristics, demographics, asset endowments, and
infrastructure among other variables. The results indicate that the income distribution
effects of the policy are very similar across different agro-climatic settings. Households
at the lowest points of the income range lose at least 25% of their incomes while the

wealthiest households either gain some proportion of their incomes or remain unaffected.

Copyright by
ELLIOT WAMBOKA MGHENYI
2006

DEDICATION
This work is dedicated to my father, the late Mzee Zakayo Mghenyi. You asked me in my
childhood whether I could go all the way and I answered in the afﬁrmative, yet you left

us before that could happen. The only consolation to me is that I am well on track.

iv

ACKNOWLEDGEMENTS

I am especially grateful to my major professor, Dr. Thomas Jayne, for his support
throughout my masters program and his continued support in the Ph.D. program. His
encouragement and understanding provided the perfect environment for the proper
completion of this work. My thanks also go to Dr. Robert Myers, the professor who
always kept his doors open for me throughout this challenging period. His insights into
the difﬁculties which I faced at different times during this work are invaluable, and were
well given. Needless to say, I found more direction and motivation to excel after every
meeting. I also want to thank the other two members of the committee - Dr. John Giles
from Economics department, and Dr. Hira Koul from the department of statistics.
Another individual who tremendously assisted me is Xue Lan, a PhD. Candidate in the
department of statistics. My chapter 4 would have been different were it not for her help
in understanding partly linear semi-parametric models.

The food security project in the department of agricultural economics has
provided me with ﬁnancial support throughout my masters program, and for that I give
my thanks. My thanks also go to Tegemeo Institute of Agricultural Policy and
Development for providing me with household survey data. Lastly, I want to express my
gratitude to colleagues in cook hall who have helped me in many ways at various times
during my stay in East Lansing. I sincerely wish you all the best out of the labors of your

work, now and in future.

TABLE OF CONTENTS

LIST OF TABLES ............................................................................... viii
LIST OF FIGURES ................................................................................. x
CHAPTER 1 INTRODUCTION .................................................................. 1

CHAPTER 2 OVERVIEW OF RURAL HOUSEHOLDS MAIZE PRODUCTION

2.1
2.2

2.3
2.4

2.6

CHAPTER 3

3.1
3.2

3.3

3.4

3.5
CHAPTER 4

4.1
4.2

4.3

AND MARKET PARTICIPATION

Introduction ........................................................................ 4
Data ................................................................................. 4
2.2.1 Sampling Procedures .................................................... 5
Comparative Farm Production and Marketing ................................ 7
Regional Patterns in Maize Production, Marketing and
Household Incomes .............................................................. 11
Conclusions ....................................................................... l6
MAIZE PRICE POLICY AND INCOME POVERTY
Introduction ........................................................................ 1 8
Conceptual Framework .......................................................... 21
3.2.1 Estimating Changes in Income due to the Maize Pricing Policy..21
3.2.2 Maize Price Elasticity of Income ...................................... 25
3.2.3 A Second Order Approximation ....................................... 26
3.2.4 Own Price Demand and Supply Elasticity for Maize ............... 28
3.2.5 Effects of Policy on the Mean of Maize Prices ...................... 29
3.2.6 Measures of Poverty ..................................................... 31
3.2.7 Stochastic Dominance and Poverty Dominance ..................... 32
3.2.8 The Setting ................................................................ 33
Data Construction .................................................................. 34
Results .............................................................................. 36
Conclusions ........................................................................ 47
MAIZE PRICE POLICY AND INCOME DISTRIBUTION
Introduction ........................................................................ 50
Conceptual Framework ........................................................... 53
4.2.1 The Semi-parametric Model ............................................ 53
4.2.2 The Estimation Strategy ................................................. 55
4.2.2.1 Estimating the Parametric Part ................................. 58
4.2.2.1 Non-parametric Regressions ................................... 60
Data Construction ................................................................. 62

vi

4.4 Results .............................................................................. 63

4.4.1 The Inﬂuence of Factors Other Than Income ........................ 63

4.4.2 Expected Changes in Income along its Range ......................... 65

4.4.2.1 Concerns about Measurement Error ........................ 71

4.5 Conclusions ......................................................................... 72
CHAPTER 5 SUMMARY AND CONCLUSIONS .......................................... 75
APPENDIX .......................................................................................... 81
BIBLIOGRAPHY ................................................................................... 85

vii

LIST OF TABLES

Table 2.1

Table 2.2

Table 2.3

Table 2.4

Table 2.5

Table 2.6

Table 3.1

Table 3.2

Table 3.3

Table 3.4

Table 3.5

Table 3.6

Table 3.7

Table A4.1

Table A4.1

Table A41

The Sample ......................................................................

Total Production and Marketing of Annual Crops:

Aggregate (1999/2000 production year) ...................................

Total Production and Marketing of Annual Crops:

Aggregate (2003/2004 production year) Constant 2000 USS ..............

Production and Marketing Patterns by Agro-climatic Zones

(1999/2000 and 2003/2004 production years) ..............................
Households Maize Marketing Position ......................................
Description of the Sampled Agro-climatic Zones ..........................

Maize Supply and Demand Elasticity .......................................

Summary of cumulative NCPB and Maize Import Tariff Effects on
Kitale and Nairobi Wholesale Maize Prices, Nominal Ksh per 90 kg

bag........ ...............................................................................................

First Order Poverty Dominance of Counterfactual Incomes

over Incomes with Effects of Price Controls: All lowlands. . . . . . . . . . ..

First Order Poverty Dominance of Counterfactual Incomes

over Incomes with Effects of Price Controls: All highlands ..............

First Order Poverty Dominance of Counterfactual Incomes

over Incomes with Effects of Price Controls: Western transitional. . . .

First Order Poverty Dominance of Counterfactual Incomes
over Incomes with Effects of Price Controls:

High potential maize zone .....................................................................

Second Order Poverty Dominance of Counterfactual Incomes
over Incomes with Effects of Price Controls:

High potential maize zone .....................................................................
The Effects of Covariates Other than Income: Coastal Lowlands .......
The Effects of Covariates Other than Income: Eastern Lowlands

The Effects of Covariates Other than Income: Western Lowlands. . . ..

viii

...7

....8

..... 9

...12

...14

..15

...29

....30

.39

...41

.44

...46

...46

..81

.81

82

Table A4.1

Table A4.1

Table A4.1

Table A41

The Effects of Covariates Other than Income: HPMZ ....................... 82
The Effects of Covariates Other than Income: Central Highlands ......... 83
The Effects of Covariates Other than Income: Western Highlands ........83

The Effects of Covariates Other than Income: Western Transitional. . .....84

ix

LIST OF FIGURES

Figure 2.1

Figure 2.2

Figure 3.1

Figure 3.2

Figure 3.3

Figure 3.4

Figure 3.5

Figure 3.6

Figure 3.7

Figure 3.8

Figure 3.9

Figure 4.1
Figure 4.2
Figure 4.3
Figure 4.4

Figure 4.5

Lorenz Curves for Pooled Production Volumes
(1999/2000 and 2003/2004) ...................................................... 10

Lorenz Curves for Pooled Gross Revenues
(1999/2000 and 2003/2004)....... ............................................... 11

Coastal lowlands: First Order Poverty Dominance Curves
(1 9 % increase in maize prices) ................................................. 37

Eastern lowlands: First Order Poverty Dominance Curves
(l9 % increase in maize prices) ................................................. 38

Western lowlands: First Order Poverty Dominance Curves
(l9 % increase in maize prices) ................................................. 38

Western highlands: First Order Poverty Dominance Curves
(19 % increase in maize prices) ................................................. 41

Central highlands: First Order Poverty Dominance Curves
(19 % increase in maize prices) ................................................. 42

Western transitional: First Order Poverty Dominance Curves
(l9 % increase in maize prices) ................................................. 43

Second Order Poverty Dominance: Western transitional:
(l9 % increase in maize prices) ................................................. 44

High Potential Maize Zone: First Order Poverty Dominance Curves
(19 % increase in maize prices) ................................................. 45

Second Order Poverty Dominance: High Potential Maize Zone:

(19 % increase in maize prices) ................................................. 46
Expected Changes in Income in Coastal Lowlands ........................... 66
Expected Changes in Income in Eastern Lowlands .......................... 67
Expected Changes in Income in Western Lowlands .......................... 68
Expected Changes in Income in Central Highlands .......................... 68
Expected Changes in Income in the High Potential Maize Zone ........... 69

Figure 4.6 Expected Changes in Income in Western Highlands ........................ 70

Figure 4.7 Expected Changes in Income in Western Transitional ...................... 71

xi

CHAPTER 1
INTRODUCTION

Maize is the most commonly grown crop in Kenya, and is the staple food for over 80%
of the population. Maize is grown in virtually all agro-ecological zones, ranging from the
Kenyan highlands to the semi-arid zones and the humid coastal lowlands. The crop
supplies 40% to 45% of the calories, and 35% to 40% of the protein consumed by an
average Kenyan. It is estimated that maize accounts for 20% of all agricultural
production and 25% of agricultural employment (GOK 2003). A large portion of maize
production (60%) is attributable to smallholders. However, large-scale commercial farms
contribute a signiﬁcant amount of total marketed maize output (Hassan and Karanja
1997; Jayne et al., 2002).

Because of its strategic position in Kenyan agriculture, maize dominates national
food security considerations. Indeed, maize pricing was the central theme during the
implementation of liberalization and privatization of key sectors in agriculture. Prior to
these reforms, Kenya had government-controlled maize trading environment and
agricultural input delivery system. The government was responsible for setting pan-
seasonal and pan-territorial maize prices, as well as prices of inputs such as seeds and
fertilizers. Maize marketing was monopolized by the National Cereals and Produce
Board (N CPB), a grain marketing board run by the government.

The maize sector reform began in 1987/88 and intensiﬁed in the 19908, when
under pressure from international lenders, the government eliminated movement and
price controls on maize trading, deregulated maize and maize meal prices, and eliminated

direct subsidies on maize sold to registered millers (Jayne and Kodhek 1997). By the end

of 1993, the market for maize was fully liberalized; maize could be distributed freely in
the country by willing traders, and imported with minimal restrictions upon payment of a
set tariff. The NCPB remained active in a liberalized market, but its role was reduced
from that of a sole trader to an agency buying maize for the purpose of building national
strategic reserves. These reforms proved controversial. Farmer lobby groups
successfully argued that lower maize producer prices were a disincentive to production
and therefore a direct threat to national food security. In 1999, the government reinstated
NCPB to purchase maize at ﬁxed support prices. The Government has since been
engaged in a three-fold intervention in the maize market; applying import tariffs,
authorizing procurement by the NCPB, and imposing non-tariff barriers on imports.
These policies are strategies aimed at "getting the prices right" i.e. bidding up maize
prices so that producers have more incentives to cultivate maize.

The rationale behind ensuring higher producer prices is based on the conjecture
that for both microeconomic and macroeconomic reasons, no country has managed to
sustain rapid economic growth without ﬁrst obtaining food self-sufﬁciency, at least in
the main staple (Timmer 1998a). At the micro-economic level, inadequate and irregular
access to food limits labor productivity and reduces investment in human capital (Bliss
and stem 1978; Strauss 1986; Fogel 1994). The macroeconomic impact of periodic food
crises is to undermine both economic and political stability, hence reducing the levels
and efﬁciency of investment (Timmer 1989, 1996; Dawe 1996). However, maize is the
main food item in most diets. Higher maize prices would therefore erode real incomes of
net buying households. In Kenya, maize expenditure accounts for over 18% of poor

household’s total income (GOK 2003). This suggests that higher prices might negate

efforts to ensure that the poor can afford adequate food. It is the well-known food price
dilemma concept ﬁrst articulated by Timmer (1986).

The purpose of this thesis is to explore the effects of maize policy on the welfare
of households in rural Kenya. The thesis is comprised of two papers. In the ﬁrst paper,
Chapter 3, I examine the effects of maize pricing policy on levels of poverty. This
analysis will summarize the effects of the policy on the welfare of households that are
considered poor, or become poor as a result of, but ignores effects on the welfare of the
non-poor who retain their status. From a policy perspective, there are questions such as
whether rich farmers beneﬁt more relative to poor farmers (or vice versa), and how those
patterns differ from one region to another. That is exactly the subject matter of the
second paper, chapter 4. Speciﬁcally, the second paper focuses on how the policy affects
the distribution of income.

Results from these analyses will be useful in informing national economic
planning, especially because current government and donor plans favor policies that are
geared towards poverty reduction and increased equality. The rest of the thesis is
organized as follows. The second chapter presents an overview of maize production and
marketing in Kenya. Chapter 3 serves two purposes; the ﬁrst is to provide the conceptual
framework used to estimate changes in incomes due to changing prices, and the second is
to estimate the impact of the policy on poverty. The effect of the policy on the
distribution of income is examined in chapter 4. Chapter 5 provides summary and

conclusions.

CHAPTER 2
OVERVIEW OF RURAL HOUSEHOLDS MAIZE PRODUCTION AND
MARKET PARTICIPATION

2.1 Introduction

This chapter provides background information on maize production and
marketing patterns in rural Kenya. The chapter is intended to provide context for the
analytical work that follows in the remainder of the thesis. First, I begin by discussing
the data used in this section and the rest of the study. Secondly, comparative production
and marketing patterns between maize and other crops are presented. In the third section,
maize production and consumption in seven regions marked by differing agro-climatic
potential are discussed together with the inﬂuence of rainfall patterns and household size.
The seven regions considered are Coastal lowlands (Kwale and Kiliﬁ districts), Eastern
lowlands (Taita-taveta, Machakos, Makueni, Kitui, and Mwingi districts), Western
lowlands (Kisumu and Siaya districts), Western transitional (Bungoma snd Kakamega
districts), High potential maize zone (Bomet, Nakuru, Narok, Trans nzoia, Uasin gishu,
upper Kakamega, and upper Bungoma districts), Western highlands (Kisii and Vihiga
districts), and Central highlands (Nyeri, Meru, and Muranga districts). Section 4 provides

a brief conclusion.

2.2 Data
The analysis uses household survey panel data collected by Tegemeo Institute of
Agricultural Policy and Development in Kenya during the years 2000 and 2004. The

primary purpose of the household surveys was to provide information for the Tegemeo

Agricultural Monitoring and Policy Analysis (TAMPA) project. Tegemeo data comprises
information on agricultural input usage and production outcomes, purchases for home
consumption, and sources and amounts of non farm incomes, among other variables. The
ﬁrst installment of the panel data was collected by Tegemeo in 1997. Subsequent surveys
in 2000 and 2004 were conducted using an improved questionnaire. In some modules,
the variables collected in the 1997 survey differ from those collected in 2000 and 2004; a
reﬂection of improved data quality. For the purposes of this work, I consider the 1997
survey to have been a learning exercise, and therefore data collected in that year will not
be used. The next section provides a discussion on the sampling procedure implemented
by Tegemeo. The discussion is based on information from the 2000 survey data

documentation and various Tegemeo working papers.

2.2.1 Sampling Procedures

The ﬁrst step in the sampling process was to obtain a list of all non-urban
divisions in the country from census data. The divisions were assigned to one or more
agro-ecological zones (AEZ) based on information from the District Development Plans
and the Farm Management Handbook. After dividing the Kenyan rural population into
its constituent agro-ecological zones, two or three divisions were chosen from each AEZ
based on their populations. Another factor that determined the choice of a division was
diversity in agro-ecological conditions. More diversity within the division was preferred
since the aim was to get a sample that represents well the varied conditions faced by
farmers in the country. The selected divisions fell in 24 districts. The divisions were

grouped into 9 broad agro-climatic zones - a hybrid of administrative boundaries and

agro-ecological zones. The Northern Arid zone — one of the 9 zones, was not visited in
2004 and is therefore not included in the two year panel. Also excluded is the Marginal
Rain Shadow zone because of a small sample size. Table 2.1 presents a summary of the
2000 and 2004 panel for the remaining 7 zones. .

After identifying divisions, the next steps were to select locations, sub-locations,
and villages in that order using a blind equal chance ballot. A list of households was then
drawn from each of the selected villages with the help of village officials. Balloting was
conducted to determine the beginning selection from the list. For example, if the ballot
picked number 10, every 10th household would be selected until the target number is
reached. Because of the purposive sampling technique employed in selecting divisions
and small sample sizes in some districts, concerns have been raised about whether the
Tegemeo dataset is representative, and if so at which level. The sampling technique was
aimed at selecting divisions between which there were variations in agro-climatic zones,
and which showed within diversity in order to ensure that the resultant sample reﬂected
the diverse conditions faced by farmers. As such, the sample represents well the diverse
farming environments and agro-climatic zones facing Kenyan farmers. Tegemeo data
can therefore be viewed to be representative at the level of broad agro-climatic zones

rather than administrative boundaries such as districts and provinces.

Table 2.1. The Sample

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Agra-climatic zone District Number of Percentage of Percentage of
households total sample zonal sample
Sampled
Coastal lowlands 78 5.9
Kwale 25 32.1
Kiliﬁ 53 67.9
Eastern lowlands ‘ 153 11.7
Taita-taveta 8 5.2
Machackos 21 13.7
Makueni 74 48.4
Kitui 17 11.1
Mwingi 33 21.6
Western lowlands 163 12.4
Kisumu 95 58.3
Siaya 68 41.7
Western transitional 156 11.9
Bungoma 44 28.2
Kakamega l 12 71.8
Western highlands 142 10.8
Kisii 85 59.9
Vihiga 57 40.1
High potential maize 375 28.6
zone
Bomet 38 10.1
Nakuru 102 27.2
Narok 24 6.4
Trans nzoia 56 14.9
Uasin Gishu 95 25.3
Kakamega 25 6.7
ngoma 35 9.3
Central highlands 246 18.7
Nyeri 99 40.2
Mem 8] 32.9
Muranga 66 26.8
Total 1313

 

 

 

 

 

 

 

Source: Household surveys 2000 and 2004; Tegemeo Institute

2.3 Comparative Farm Production and Marketing

The results from household surveys in 2000 and 2004 show that maize was the
leading crop enterprise in terms of farm production value (table 2.2 and table 2.3). In
both years, the percentage of households selling maize was second only to those selling
fruits and vegetables; implying that the most common staple crop is also an important

source of cash income to rural households. Similarly, fruits and vegetables rank higher

than maize in terms of gross sales value. However, horticulture is typically both capital
and labor intensive, and therefore after these factors are paid, net revenues from maize
could be higher. From the national outlook we learn that maize and horticulture are the
two leading crop enterprises, and therefore further discussions in this section will focus

on the two.

Table 2.2. Total Production and Marketing of Annual Crops: Aggregate (1999/00

production year)

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

CROP TOTAL VALUE OF % OF TOTAL SALES SALES
PRODUCT] PRODUCTI HHS SALES VALUE As %
ON (‘000 ON (2000 SELLIN (‘000 (2000 OF
KG) USS) G KG) USS) PRODU
CTION
“Tea" (excludmg 495 99,843 16 383 76,265
ma1ze) 77
Mme 2,417 359,316 38 1,151 164,948 48
Beans/Groundnuts
IOilseeds 279 75,249 30 88 23,850 31
Roots/1‘ "be“ 1,089 86,562 31 446 36,073 41
Non-food Cash
Crops 12 10,001 4 9 6,843 79
Industrial sugarcane 7,015 154,224 10 7,015 154,224 100
Tea 756 198,139 13 752 197,043 99
Coffee 268 69,942 20 262 68,399 98
Fruits and
Vegetables 4,034 268,673 74 2,107 152,143 52
Coconut and
Cashewnm 152 7,109 3 101 4,539 67

 

Source: Jayne et al 2005

 

Table 2.3. Total Production and Marketing of Annual Crops: Aggregate (2003/04
production year)

 

 

 

 

 

 

 

 

 

 

 

CROP TOTAL VALUE OF °/o OF TOTAL SALES SALES AS
PRODUCT] PRODUCT] HHS SALES VALUE °/o OF
ON (‘000 ON (2000 SELLl (‘000 KG) (2000 PRODUCT I
KG) USS) NC U55) ON
Cereals
(excluding 522 83,459 7 17 375 60,167
maize) 72
Mm" 2,353 271,503 44 1,149 132,848 49
Beans/Ground
“‘5 293 67 259 36 102 23 573
IOilseeds ’ ’
. .. 35
R°°tsnubers 1,015 63,864 35 441 28,704 43
Other Non-
food Cash 8 5,120 6 8 4,680
Crops 95
I“"“““‘"" 4,037 69,885 4 3,998 69,208
sugarcane 99
Tea 809 142,371 13 804 141,687 99
Coffee 169 13,666 18 162 17,813 96
Fruits and
Vegetables 3,581 21 1,470 78 L986 1 19,593 55
Coconut and
Cashewnut 83 2,675 4 48 1,466
57

 

 

 

 

 

 

 

 

 

Source: Jayne et a1 2005

In Kenya, fruits and vegetables have been viewed as an alternative for maize to
farming household’s with low land endowment and surplus labor. Figure 2.1 and ﬁgure
2.2 respectively provide production output and gross revenue lorenz curves for maize
and fruits and vegetables. In both crop enterprises the higher 10% of sampled households
produce 50% of total output (Figure 2.1). However, gross revenues from fruits and
vegetables are relatively spread out among more households compared to maize. Figure
2.2 shows that 75% of gross revenues from maize accrue to 10% of the sample, while for

fruits and vegetables the same proportion of the sample receives 65% Of gross revenues.

This suggests that more farmers stand to beneﬁt from cash incomes due to
commercialized horticulture than to maize.

Concentration in the distribution of gross maize revenues suggests that the maize
pricing policy might afford disproportionately large welfare gains to the 10% that sell
75% of total marketed maize. Other factors that would determine the net effects are
consumption levels and initial levels of income. The next section considers maize

production, marketing, and income levels across different agro-climatic settings.

Figure 2.1. Lorenz Curves for Pooled Production Volumes (1999/2000 and 2003l2004)

Production Volumes Lorenz curves for llalze and Fruits and Vegetables

 

1%

0.9 >
0.8 1*
0.70

0.6"

0.5‘L

L191

04>

0.3"

0.2"

om-

 

 

 

o, : . 4 E e : + : : : c : 4. a : : : : c
0.0) 0.05 0.10 0.15 0.20 025 0.30 0.35 0.0 0.45 0.50 0.55 0.50 0.05 0.70 0.75 0.80 0.05 one 0.05 1.00
Peroertlesm)

 

 

L—erﬂyotmolzeproaned -------- Wotmwvegaauesproamj

10

Figure 2.2. Lorenz Curves for Pooled Gross Revenues (1999/2000 and 2000/2004)

Gross Revenue Lorenz curves for Ilalze and Fruits and Vegetables

 

1.0T

0.9 '

0.3-
oz-
0-11’

 

O. : 4 4 4. 1‘ . . : t : a : 3 t 3 J. ‘. ﬂ :
0.00 0.05 0.10 0.15 0.20 0.25 0.30 0.35 0.43 0.45 0.50 0.55 0.00 0.05 0.70 0.75 0.00 0.05 0.90 0.05 1.00
Percerllles (p)
[— Gross revenue from maize -------- Gross revenue tram trults and vegetables I

 

 

 

 

 

2.4 Regional Patterns in Maize Production, Marketing and Household
Incomes.

Table 2.4 shows that with the exception of the high potential maize zone,
households in other agro-climatic regions spend more money on maize purchases than
the cash income they receive from maize sales. The high potential maize zone (HPMZ) is
marked by excellent maize growing conditions as summarized by rainfall levels and
altitudes in table 2.6. Maize production in the HPMZ is three times higher than in most
of the sampled agro-climatic conditions, and land holding sizes are above the average.
Jayne et a1., (2002) observe that a large portion of total marketed maize in Kenya is
produced in the HPMZ. On the other hand, households in coastal lowlands and western

lowlands incur the highest level of negative balances from maize trading. It is expected

11

that increases in the price of maize would significantly increase maize expenditures in
these regions relative to the rest of the sample. Further, incomes are lowest in the western
lowlands suggesting that farmers in these regions are faced with high maize budgets that
would exacerbate welfare losses from a maize) price increase. The western transitional
zone and the Kenyan highlands (central highlands and western highlands) portray
intermediate levels of trade imbalances from maize. Household sizes are fairly stable

across different agro-climatic zones.

Table 2.4. Production and Marketing Patterns by AgrO-climatic Zones (1999/2000 and
2003/2004 Production years)

 

 

 

 

 

 

 

 

 

 

 

Maize Production Sales value less Household Income Family size
(KG) purchases value (2000 US S) (adult equivalents)
(2000 US 3)
Production 1999/20 2000/20 1999/20 2000/20 1999/20 2000/20 1999/20 2000/20
year 00 O4 00 O4 00 O4 00 04
Medians
Coastal
Lowlands 515 360 (143) (75) 1,111 1,055 7.50 7.40
Eastern
Lowlands 745 843 (54) (12) 1,370 1,346 5.90 5.00
Western
Lowlands 270 294 (56) (30) 444 520 5.20 4.50
“/3979 810 1,047 (19) 0 1,969 880 6.00 5.53
Trans1t1onal
High
Potential 1,620 2,340 47 106 1,549 1,458 6.00 5.34
Maize Zone
Western
Hi hlands 630 786 (15) (2) 975 786 5.85 5.00
Central
Hi hlands 495 565 (13) (8) 1,882 1,342 5.50 4.00
___g* .
Na‘mm' 720 859 (16) (2) 1,365 1,139 5.90 5.00

 

 

 

 

 

 

 

 

 

 

 

Thus far, analysis in this section has focused on levels of maize sales and maize
purchases relative to income levels. The average net purchase figures presented in table

2.4 above might give an impression that all households participate in the market both as

12

buyers and sellers. That is not the case. Table 2.5 shows that in all regions, less than a
third of households participate in the market both as buyers and sellers. ‘Buyers only’ are
the majority in all zones except the high potential maize zone. This group forms up to
90% of households in coastal lowlands, and more than half in other regions. In the high
potential maize zone, the majority of households (almost half) participate as ‘sellers
only’. However, even here a substantial number of households are ‘buyers only’ (about
20%). Autarkic households are barely more than 10% in all Of the seven zones
considered.

An examination based on net purchases and net sales patterns ignores the welfare
effects of the maize pricing policy with regard to production consumed at home. Maize is
the most commonly grown crop, and is the main staple for all rural households in Kenya.
A maize price increase represents increased income with regard to the part of production
not traded. On the other hand, a welfare loss is incurred when this portion of output is
consumed; loss of consumer surplus due to increased prices. The welfare loss due to
consumption would offset the welfare gain due to increased value of production if the
proportion of production consumed at home is assumed to be inelastic to changing
prices. If this were true, then examining trends in net sales would sufﬁce for this study.
However, it is very unlikely that the proportion of production consumed at home is price
inelastic. For example, increased prices might provide incentives for hitherto net buyers
to reduce the proportion of production consumed at home and become autarkic or even
net sellers. Similar responses might occur among autarkic households who would

become net sellers, and net sellers who might want to advance their status.

13

In summary, changes in the proportion of production consumed at home are

possible for households in autarky, and for those that participate both as buyers and

sellers; in total about 30% of sampled households. The next chapter presents a

conceptual framework to examine welfare effects of households both as producers and

consumers, and as traders of maize without assuming that the proportion of production

consumed at home is price inelastic.

Table 2.5. Households maize marketing position

 

Maize marketingmsition

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Agro-climatic zone buyers only sellers only both buyers and neither buy nor
(%) (%) sellers (%) sell (%)
1999/2000 cropping season
Coastal lowlands 90 4 5 1
Eastern lowlands 68 6 21 6
Western lowlands 69 3 18 10
Western transitional 51 17 25 7
iigh potential maize zone 25 4O 23 12
Western highlands 55 19 19 7
Central highlands 58 l l 23 8
Total in 2000 55 17 2O 8
20003/2004 cropping season
Coastal lowlands 88 O 9 3
Eastern lowlands 50 13 27 10
Western lowlands 73 6 15 6
Western transitional 37 24 27 12
High potential maize zone 19 49 24 8
Western highlands 43 22 23 12
Central highlands 58 l l 18 13
Total in 2004 50 16 26 8

 

14

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

. . . . 852$:
t: 6 8m 0 mom 5. «8 «No 3. 80 «8 8w mam 32.8
. . . . €529:
Rm m x: 6 «mm m. :a Sm o8 Sm 9.0 F 3» F2 5283
BEN 332
$06 83 Sam 8m. mum m3 3» :5 new 3.58..
an 2.2:
. . . . . 230335;.
So 4 3m m mam v cum So P .5 £3 E F :5 98 5283
. . . 8533
mg m one 4 v8 m E. PR 5o mos New mac t; 5383
. . . €5.33
80 4 «8 m 9: m 3:. mt. 8m 2: «R m: 3: seem
85.33
E. Km 2 SN 8v mu? «.2 mm: mm 2. .988
:32 835582 52552 :32 Esﬁsaz EEEEZ 502 825582 EsEEm—z
aaoséé aegis: Baeam
20>». new o>onm anezv a?» 5:03.05 8833— 39» cote—625 vcchoom mEosomso:
ovam=< ”6030.9. 59.: 05 E =au£am ”Gamma Ea:— 05 E =ﬂ£m¢ no hoop—52.

 

neneN 53.555.333.558 95 we 523.839 6d ozah

15

2.5 Conclusions

This chapter provides an overview of maize production and consumption patterns
in rural Kenya. The purpose of the chapter is to give context to the analysis in later
sections of the thesis. Most of the conclusions may not be new, nevertheless; I will
brieﬂy discuss key observations.

Maize is the largest source of crOp incomes for rural households in Kenya. Even
though the crop is an important food staple, the percentage of households engaged in
maize selling is second only to those selling fruits and vegetables. Also, gross revenues
from maize are only exceeded by horticulture. This means that the most important staple
is also a vital source of cash income to farmers. The pattern of maize production and
marketing varies greatly between regions with differing agro-climatic conditions. On
average, rural households purchase maize worth more than their cash earnings from
maize sales. The highest trade imbalance from maize was observed among farmers in
coastal lowlands and western lowlands. Further, up to three quarters of gross revenues
from maize accrue only to a tenth of households sampled. From these preliminary
ﬁndings it is clear that the price policy on maize will have differing impacts on
households faced with different agro-climatic conditions. As such, analysis of welfare
effects at the national level will mask important variations across regions. I therefore
adopt broad agro-climatic zones as the level of analysis.

An analysis solely based on sales and purchase values ignores welfare effects
with regard to the portion of production consumed at home. Such an approach would
suffice if the proportion of production consmned at home were assumed to be inelastic to

changes in the price level. The assumption may appear plausible given that the crop is

16

the main staple in diets of farming households. However, maize is also a source of cash
income, and therefore it is expected that higher maize prices could make selling more
attractive while stocks last, and prohibit purchases when stocks are depleted, thereby
reducing the proportion of production consumed at home. Price responses that change
the proportion of production consumed at home are possible among households that
participate both as buyers and sellers, and those in autarky. The two groups form about
30% of households in the sample - a substantial proportion. In the next chapter, a
conceptual framework that allows for price responsiveness in the proportion of
production consumed at home is presented. A second order approximation of equilibrium
price changes discussed in section 3.2.3 forms the basis for estimating welfare effects of

the price policy.

17

CHAPTER 3

MAIZE PRICE POLICY AND INCOME POVERTY

Abstract
This chapter serves two purposes. The ﬁrst is to provide the conceptual framework used
in estimating changes in incomes due to the price policy, a discussion that leads to the
second order approximation of equilibrium income changes. Secondly, based on the
framework, poverty orderings are generated between two income distributions;
counterfactual incomes and incomes with the effect of price controls. Results indicate
that the policy has aggravated poverty in all regions considered, except the high potential
maize zone. Under the price policy, the number of the poor is higher in the Kenyan
lowlands and the highlands. The policy does not increase the number of the poor in the
western transitional zone; however, their income shortfall from the poverty line is

increased.

3.1 Introduction

The importance of food price levels on the welfare of producers and consumers
has over the years led governments to consider ‘getting prices right’ especially on key
food commodities. The ‘food price dilemma’, articulated in Timmer (1986) embodies
conﬂicting interests between producers and consumers of food commodities. Timmer’s
seminal contribution has been followed by an array of empirical work mostly in the
context of developing countries. Empirical studies on the subject include; Deaton (1989)

on rice prices and income distribution in Thailand, Barrettt and Dorosh (1996) on

18

changing rice prices and farmers welfare in Madagascar, and Budd (1993) on changing
food prices and rural welfare in Cote d’I Voire.

The main analytical tool used in these studies is the ‘net beneﬁt ratio’ or ‘net
consumption ratio’ that is used to estimate instantaneous (ﬁrst order) welfare effects of
changing prices‘. By considering both supply and consumption adjustments together with
the attendant responses in the rural wage labor markets, this paper extends the ﬁrst-order
model used in previous studies and considers a second order approximation of
equilibrium welfare changes. Speciﬁcally, the chapter begins by providing a conceptual
framework that leads to the development of the maize price elasticity of income, akin to
the ratio used in previous work. The elasticity is then used to incorporate own price
effects on maize supply and demand together with accompanying adjustments in rural
wage labor market to estimate a second order approximation to equilibrium welfare
changes.

Previous studies have inferred poverty outcomes from analyses of income
distributional effects, probably due to a perceived correspondence between the two
concepts. My hypothesis is that it is possible to have a policy that raises the incomes of
the very poor yet causes more poverty; for instance, if the welfare gains are not enough to
pull the beneﬁciaries out of poverty and are borne out of transfers from some of the non-
poor who become poor. With this understanding, the effect of the policy on poverty
examined here is separated from analysis on the effect of the policy on the distribution of
income, which is the subject matter of chapter 4.

Most poverty analyses use some speciﬁed poverty line. The weakness of this

approach is that when the threshold is varied, poverty rankings may be reversed which is

 

' The ratio is calculated as production value less sales value as a ratio of expenditure

19

clearly undesirable in the context of policy analysis. In this paper, methods that generate
results which are robust to a wide range of poverty lines are used. First, the study
considers all magnitudes of price changes within 5% standard deviation of the price
impact suggested by (Jayne et a1., 2005a). Each simulated price change is used to
generate two vectors of incomes - counterfactual incomes and incomes that have occurred
with price controls. Next, the paper exploits the parallel between stochastic dominance
and commonly used measures of poverty to generate poverty orderings from each pair of
income vectors. The poverty rankings generated hold for a wide array of poverty lines,
including the World Banks $1 a day per person threshold and a poverty line for rural
households developed and used by the government of Kenya. DAD2 software is used to
conduct stochastic (poverty) dominance tests. The software has previously been used for
multidimensional poverty analysis in (Duclos et a1., 2005) and in a study on Indonesian
living standards (Strauss et a1., 2004).

Results indicate that NCPB price supports and tariffs on imported maize jointly
plunge some households into transitory poverty in all regions considered except the high
potential maize zone (HPMZ). The proportion of the poor is increased in coastal
lowlands, eastern lowlands, western lowlands, western highlands, and central highlands.
In the western transitional zone, the policy may not have led to an increase in the number
of the poor; however, their income shortfalls are increased. These results hold even when
both supply and demand responses are assmned to differ across income ranges.

This chapter is organized as follows; the next section provides the conceptual

framework for estimating changes in income due to the maize price policy, a discussion

 

2 The software is well tailored for distributive analysis. It can be obtained upon request from its author
Jean-Yves Duclos at www.ccn.ulaval.ca/~jyves/

20

that leads to a second order approximation of equilibrium changes in welfare. A note on
measures of poverty and the parallel between stochastic dominance and poverty
dominance completes the section. Section 3 brieﬂy discusses data construction and

section 4 presents the results. Conclusions are given in section 5.

3.2 Conceptual Framework
3.2.1 Estimating Changes in Income due to the Maize Pricing Policy

A social welfare function is postulated in theory as a weighted aggregate of
individual utility functions. In reference to differing deﬁnitions and views on the social
welfare function (Varian 1992) asserts that the most reasonable interpretation of such a
function is that it represents a social decision maker’s preference about how to trade off
utilities of different individuals or households. According to (Deaton 1997), the social
welfare function should not be thought of as the objective ﬁrnction of a government or
policy-making agents, because there are few if any countries for which maximization of
some social welfare function subject to constraints would provide an adequate description
of the political economy of decision making. Instead, the social welfare function "should
be seen as a statistical aggregator that turns a distribution into a single number which
provides an overall judgment on that distribution, and that forces us to think coherently
about welfare and its distribution". While economists hold varying views on practical
ramiﬁcations of the social welfare function, the nexus of agreement is that a social
welfare function can be used to theorize the distributional principles underlying economic
policy.

In Deaton (1989), the indirect utility function of a household is given as

21

V=w(wT+b+7r,p). (3.1)
Where, V is utility value of household i, w is the wage rate, T is the total time worked,
b is rental income or transfers, p is price vector, and 7: is the household’s proﬁts from
farming or other family business. Households are assumed to be proﬁt maximizers and
therefore 7: is the value of the proﬁt function 7r( p, u, w ). Where, u is a vector of input
prices and w is wage rate.

Without attempting to speculate about functional form and assuming short run
proﬁt maximization decisions on rental income and wage employment, the representation

of the indirect utility function can generally be given by
Vl7r(pp,u,w), pCJ. (3.2)

Where p p and pc are vectors of producer prices and consumer prices respectively. The
price vector pp includes rental property rates. Similarly, u includes prices associated with

maintenance of rental property together with input prices in farm production and family
businesses. Further, the price of maize ( pm) can be separated from the vector of other
prices to give,

Vlﬂ(pm, pp,u, w), pm, 1),]. (3.3)
For simplicity, the study does not distinguish between consumer and producer prices for
maize. This approach is tenable when consumer and producer prices are assumed to
change by the same proportion. Equation (3.3) is the utility value of an individual
household. The aggregate welfare level is the value of a social welfare function that
aggregates household utility levels in accordance with a social decision maker’s

distributional principles.

22

W=26V. (3.4)

1 :
(=1

The 6 parameters are subjective and represent hypothetical weights that the policy maker
attaches to utilities of different households. In section 3.2.7 variants of Equation (3.4)
namely; the Foster Greer and Thorbecke (FGT) measures of poverty will be used to
assess aggregate welfare effects of a change in the price of maize.

To study the effects of a change in the price of maize on a household’s level of
utility we begin by totally differentiating Equation (3.3) holding all variables other than

the price of maize constant.

dV = (0%” * 6%1). )dpm + 59%;)”, dpm (3.5)

From Hotelling’s lemma

an _
5pm — 0",. (3.6)

Where am is the proﬁt maximizing maize output. Roy’s identity implies that

6%“ = -c.. 5%”. (3.7)

In Equation (3.7), cm is the utility maximizing quantity of maize consumed. Substituting

Equation (3.6) and Equation (3.7) into Equation (3.5) and re-arranging terms we get
dV=5V/a,,(o. —c.)dp.. . (3.8)
The ﬁrst component of Equation (3.8) 5%” is the marginal utility of income (or

marginal utility of households’ proﬁt). The second component (0", - cm )dpm is the change
in net income caused by the change in the price of maize. The product of the two gives

the change in utility due to a change in the price of maize. Essentially 5%” is a

23

transformation factor that maps changes in incomes into utility changes. The second
component will be computed from the 2000 and 2004 household survey data.
To estimate the marginal utility of income one would require an explicit model of

the supply and demand systems. However, limitations in data availability render such

estimation infeasible and therefore instead of estimating 6%”, we assume that the

marginal utility of income is constant across all households.3 With this assumption

5%” is a common scaling factor that can be standardized at a value of 1, which is

equivalent to assuming that changes in incomes are fully transformed into utility changes
in a one-to-one correspondence for all households. In this manner, income has an
interpretation as a ‘money metric of utility’. The problem with using income as a ‘money
metric of utility’ is that a certain level of it does not purchase the same basket of goods in
different regions, more so because of geographical differences in prices. We therefore use
same input and output prices in measuring household income, as well as same consumer
prices for maize grains and maize products nationally.

From Equation (3.8), the realization of a change in utility for household i is
W. = (0., —c... )dp. = dy.. (3.9)
In the very short run 0,," and cm, are assumed to be fairly stable because of biological

production lags and considering that the crop is the main staple in the diets of most

households in Kenya.

 

3 This is equivalent to one of the fundamental assumptions in demand analysis; that aggregate demand is a
function of prices and aggregate wealth. Both are obtainable when individual preferences admit indirect
utility ﬁmctions of the Gorman form i.e v, (p,w,) = a,(p) + b(p) w, . For further discussions on this topic
Mas-colell et a1 (1995) and Deaton and Muellbauer (1980) are good references.

24

3.2.2 Maize Price Elasticity of Income

When divided by incOme before the price change, Equation (3.9) becomes

dyr/y10=(qm1—lmr)dlnpm' (3'10)

Where, qm, =(p; xom,)/y,° is the value of maize production (gross revenue) for

household i as a proportion of household income, and I,“ = (p; xcm,)/y," is the budget

share of maize. Equation (3.10) can be re-arranged to give

d,/»,° _ _
(y ))/(dlnpm)_(qm I...) (3.11)

Since (dy, /y;’ )= d ln y, , Equation (3.1 1) becomes

d In y, _ _
Alf! pm — (qnu [m1)' (3‘12)

Equation (3.12) can readily be interpreted as the maize price elasticity of income. The
elasticity is positive for households whose gross incomes from maize are larger than their
budgetary expenditure on the commodity. Households purchase maize in a variety of
forms ranging from whole maize grains to siﬁed maize meal. Therefore, the budget share
of maize is comprised of multiple products:

1.. =20); xcj) yr. (3.13)

J
Where, each j is a different form of maize purchased.
The very short run (VSR) percentage change in per—capita income is computed

thus

VSR = (qrm —[m1ypm-percenl ' (314)

25

Where, dpmpcmn, is the percentage change in the price of maize. The maize price

elasticity of income in Equation (3.12) is analogous to the ‘net beneﬁt ratio’ (NBR) or
‘net consumption ratio’ used to study the impact of food price changes on income
distribution in Deaton (1989), Budd (1993), and Barrett and Dorosh (1996). These studies
employed models that focused on the instantaneous effects of price policy on real
incomes. In this study, the analysis is extended to include second round effects that
account for both the response of households as demanders of the staple and a supply

response, thereby providing an approximation to equilibrium responses to price changes.

3.2.3 A Second Order Approximation

Consumers are expected to respond to a price change as time passes, while a
supply response would likely occur in the next production cycle. In areas with two
cropping seasons, the supply response could be felt within the same year, while in the
main grain basket of Uasin Gishu and Trans Nzoia districts, the response will happen a
year later or longer because of a single cropping season. The supply side response will be
accompanied by adjustments in rural wage labor markets for maize. Assuming under-
employment of rural labor, the short run effects of supply adjustments are not likely to
change the wage rate; rather the wage bill will change proportionally to the supply
elasticity due to changes in man-hours hired. Under these postulates, the complete

second order approximation of changes (SOAC) in income is given by

SOAC = (q... -I... )dp..-,...... + % [(q... )4. - (I... :1. lemma)?

(3.15)
+ V2 [85... (WSW - wrm. )kdpm-,,.,..,., )2

26

Where, 5;: is own price elasticity of maize supply, a” is the own price elasticity of

m:

maize demand, wsm, is the share of income from hired-out farm labor associated with the
maize enterprise, and wrm, are payments to hired labor for maize production as a

proportion of income. Further economy-wide adjustments are expected, especially in the
production and consumption of compliments and substitutes.

The ﬁrst part of Equation 3.15 is the percentage change in income evaluated at the
initial share of maize income and initial budget share - the very short run effects. The
second and third parts comprise the remainder term evaluated at some point between the
initial values and the unknown equilibrium values of maize income share and maize
budget share. We assume that at that point, there would have been sufﬁcient adjustments
such that the maize income share and budget share will approximate their initial values,
even though production, consumption, and income levels would have changed.
Therefore, the second derivatives with respect to income shares and budget shares are
evaluated at their respective initial values; and so are hired labor payments to maize and
income shares from supplying labor to maize farms.

The remainder term is an approximation of higher order effects after economy-
wide adjustments on markets for other commodities and rural farm wages. Such impacts
can ideally be estimated with the aid of a Computable General Equilibrium (CGE)
models. However, standard CGE models based on household data would involve
considerable aggregation across household types. Chen and Ravallion (2004) contend
that they form crude tools for welfare distributional analysis and therefore do not yield
results that can be considered necessarily superior to those from second order

approximated equilibrium conﬁgurations.

27

The very short run (ﬁrst order) effects are instantaneous income changes that are
embedded in the income data of households. Therefore, to generate a second order
approximation of resultant income changes only the second and third parts of Equation
(3.15) are applied on income data. On the same vein, counterfactual incomes are
generated by subtracting the ﬁrst part of Equation (3.15) from the household income data.

The policy is estimated to have increased the mean of maize prices by 19.7% between

1995 and 2004 (Jayne et a1., 2005). Therefore we consider 19.7% for dpmwm, in

Equation 3.15. We also consider 15% and 25% so that we can determine the sensitivity of
impacts to the degree of price change, which might in fact vary somewhat geographically

across the sample.

3.2.4 Own price demand and supply elasticity for maize.

A number of studies have estimated supply and demand elasticity of different
categories of commodities. Even when similar methodologies were used the range of
estimates differed (Rao 1988). Rao (1988) observes that a critical review of the literature
reveals that cross sectional estimation of supply elasticity exaggerates aggregate supply
responsiveness to prices, while time series studies underestimate the response somewhat.
According to Chhibber (1989), the ranges of supply elasticity for developing countries
could vary from as low as 0.2 for countries with very poor infrastructure to about 0.9 for
countries with advanced infrastructure. Karanja (2003) summarizes short run supply and
demand elasticity by zone from (Munyi 2000) and (Pitt and Sumodiningrat 1991). In the
summary, the elasticity for different zones fall within the boundaries suggested by

Chhibber (1989), and are therefore considered plausible for the purposes of this study.

28

Table 3.1 adapted from Karanja (2003) gives the following elasticity’s for different

zones.

Table 3.1. Maize Supply and Demand Elasticity

 

 

 

 

 

 

 

 

Zone Own price supply elasticity Own price demand elasticity
Lowlands 0. 199 -0.400
Dry mid-altitude 0.399 -0.400
Moist mid-altitude 0.305 -0.450
Dry transition 0.338 -0.450
Moist transition 0.378 -0.500
Highlands 0.326 -0.500

 

 

 

Source: Karanja (2003)

3.2.5 Effects of Policy on the Mean of Maize Prices

The twin polices of NCPB price support and tariff on imports have had varying
intensity since liberalization began in the late 19805. During these early years of
liberalization through early 19905, NCPB was given substantial ﬁnancing which was
enough to purchase between 3 and 6 million bags per year - more than a half of
domestically marketed maize output (Jayne et a1., 2005a). By the mid 19905, the
government had reduced the NCPB operating budget, which resulted in reduced
purchases of about 1 million bags during the 1994/95 marketing year. The government
reneged on these reforms in 1999 due to intensive lobbying by commercial maize farmers
and reinstated NCPB to purchase maize at support prices. Tariff on imported maize has
also been varying over the years. The highest rate attained over the years was 50%.

Between mid 1992 and mid 1995 the tariff on imported maize was scrapped altogether.

29

 

Despite the importance of maize prices in determining real incomes of both rural
and urban households in Kenya, no study has been done to estimate the effect of NCPB
purchases and import tariffs on mean levels and volatility of maize prices until Jayne et
a1., (2005a). They use a VAR technique to assess both the separate and joint impacts of
the twin policy on wholesale prices in Kitale, a maize surplus region and Nairobi, an
urban center with consumers in excess 'of 3 million. Table 3.2 below gives a summary of

their ﬁndings.

Table 3.2. Summary of Cumulative NCPB and Maize Import Tariff Effects on Kitale and
Nairobi Wholesale Maize Prices, Nominal Ksh per 90kg_ba

 

 

 

 

 

 

 

Kitale wholesale Nairobi wholesale
Period maize price maize price
(Ksh per 90kg bag) (Ksh per 90kg bag)
Historical No % Historical No %
NCPB difference NCPB difference

April I 990 - May I 992

Mean 335.24 227.81 47.6% 422.58 320.40 31.6%

Standard deviation 101.61 101.04 0.6% 53.77 86.73 -38.0%

Coefﬁcient of variation 30.3% 44.5% -31.8% 12.7% 27.1% -52.9%
June 1992 — June I 995

Mean 723.20 91 1.41 -20.7% 898.76 1081.20 -16.9%

Standard deviation 216.83 349.32 -37.9% 164.34 352.23 -53.3%

Coefﬁcient of variation 30.0% 38.33% -21.8% 18.3% 32.58% -43.9%
July I 995 — June 2004

Mean 996.71 831.79 19.7% 1212.56 1027.33 18.0%

Standard deviation 308.09 347.84 -11.4% 277.27 371.91 -25.4%

Coefficient of variation 30.9% 41.8% -26.0% 22.9% 36.2% -36.8%
Overall sample period
(April I 990 —- June 2004)

Mean 848.56 787.83 7.7% 1033.64 963.19 7.3%

Standard deviation 355.34 397.69 -10.6% 365.58 429.33 -14.8%

Coefﬁcient of variation 41.9% 50.5% -17.0% 35.4% 44.6% -20.7%

 

 

 

 

 

 

 

 

 

 

Note: “Historical” refers to actual prices; “no NCPB” prices are simulated from model results.
Source: Jayne, T. 8., R. J. Myers, and J. K. Nyoro (2005a).

The study ﬁnds that between July 1995 and June 2004, the twin policy was

responsible for a 19.7% increase in the price of maize in Kitale. Nairobi maize prices

30

were increased by 18%. In addition to the price impacts found in Jayne et al., (2005a),
this study also considers increases of 15% and 25% so as to ensure results that are robust

to variations in the magnitude of changes across regions.

3.2.6 Measures of Poverty

When a social welfare function such as that postulated in Equation (3.4) gives
zero weight to the welfare of households that are well off, it becomes a measure of the
welfare of the poor (Deaton 1997). Viewed this way, poverty measures are special cases
of social welfare functions; their distinction being the use of some poverty line as a
yardstick to decide between utilities that are given positive weights and those that get
zero weights. The use of a poverty line suggests some discontinuity of welfare with
poverty on one side and lack of it on the other side. Existence of such a discontinuity in
commonly used indicators (income, consumption, calories e.t.c.) has been doubted
severely by many writers. Nevertheless, one can rationalize a poverty line by considering
it to be the minimum expenditure needed to attain some threshold level of utility at given
reference prices.

To quantify poverty the study employs the commonly used Foster, Greer, and
Thorbecke (FGT) decomposable measures of poverty (Foster et a1., 1984). Considering a

cumulative income distribution function F ()2) that is right continuous and non-

decreasing, Foster and Shorrocks (1988a) give the FGT poverty measures as

 

Pa(F;z)= 2:, [(z)[z—F"'(p)]a—ldp, a2] (3.16)

31

Where y = (y1 , y2, y3 , ........ y") is a vector of per capita household incomes in increasing
order and z is the poverty line. The headcount ratio p, (F ;2) measures the proportion of

households falling below the poverty line and the income gap measure p2 (F ;2)

considers total weighted shortfalls of individual household incomes below the poverty
line - the weight being the poverty line itself. Welfare evaluations based on the FGT
measures of poverty usually hinge on the choice of a particular poverty line. As such, it is
possible that welfare rankings between income distributions are reversed with a different
choice of a poverty line - an ambiguity that is clearly undesirable. The literature refers to
this issue as the identiﬁcation problem. In the next section, I will describe an approach
that effectively addresses the identiﬁcation problem by giving poverty rankings that are

robust to a wide range of poverty lines.
3.2.7 Stochastic Dominance and Poverty Dominance

Foster and Shorrocks (1988a, 1988b) demonstrate that the PCT poverty measures
correspond to a stochastic dominance partial ordering. Essentially, the poverty ordering

Pa is precisely the stochastic dominance ordering denoted Da . If Pa is the measure of
poverty, we say F (2) has more poverty or at least as much poverty as 0(2) if 0(2)
dominates F (2) in the a degree. For the purposes of this work, pl (F ;2) the headcount
ratio, and p2 (F ;2) the income gap measure are used; implying that ﬁrst order D1 and

second order D2 stochastic dominance evaluations are respectively considered.

A distribution 0(2) dominates another F (2) in the ﬁrst degree if the value of its

cumulative distribution is less than or equal to that of F (2) for all 2 , and strictly less

32

than that of F (2) in at least one 2, . This would imply that F (2) has a higher probability
for lower values compared to 0(2) , and therefore the headcount ratio is higher in F than

in 0 for any poverty line in (0,2]. Second degree poverty dominance could be used

whenever ﬁrst degree dominance cannot be established. This is equivalent to using the

poverty gap measure when the headcount ratio fails to give conclusive results.

A distribution 0(2) dominates another F (2) in the second degree when the

cumulative difference of the area under F (2) from the area under 0(2) is non-positive,
formally given as I [0(y)— F (y)]dy S 0 for yE[0,2]. 0(2) dominates F (2) in the second

degree translates to the conclusion that the headcount ratio may be the same in 0 as in

F but income shortfalls from any poverty line in (0,2] are higher in F than they are in

0.

Previous studies that have used dominance approaches include a comparison of
multidimensional poverty ordering in Ghana, Madagascar, and Uganda (Duclos et a1.,
2005) and an analysis of Indonesian living standards (Strauss et a1., 2004). However, this
is the ﬁrst application of these methods to examine welfare impacts of changing food

prices.
3.2.8 The Setting

One of the major drawbacks of household surveys is that samples are often not
representative at smaller disaggregated levels such as villages, and therefore estimates at
these levels fail to have good statistical properties (Ghosh and Rao 1994). This issue has

led to the development of poverty mapping techniques that combine household survey

33

data with census data to develop small area estimates of poverty. The method takes
advantage of the detailed information from household surveys and the wide coverage of a
census to develop estimates for small areas. It is based on literature in small area
estimation discussed in Ghosh and Rao (1994) and Rao (1999). Applications to welfare

estimation are developed in Elbers et a1., (2003).

This paper does not apply poverty mapping techniques because they are not
needed where broad agro-ecological zones are the most suitable level of analysis.
Moreover, for policy considerations the interest will hardly be on small geographical
units such as villages, towns, or administrative classiﬁcations. Rather, policymakers
would be interested in understanding welfare effects on regions characterized by differing
maize productive potential, agro-ecological conditions, and input type usages. This way,
same estimates are generated for regions within which rainfall, topography, market

infrastructure, seed material usage, and soil types are fairly stable.

3.3 Data Construction

Poverty dominance analysis will require two vectors of income - counterfactual
incomes and incomes with effects of price supports. These variables are constructed from
the data described in section 2.2. I ﬁrst compute incomes from various sources at
household level. The sources are; crop income, livestock income, salaries and
remittances, and incomes from business and informal activities. These are summed up to
give total household income. The computations are done separately on the 2000 and
2004 survey data to generate a panel data of total household income. The vector of 2004
total household income is then deﬂated to 2000 using the relative ofﬁcial inﬂation rate

proﬁle between the two years.

34

The next step is to use the pooled total household income and generate the two
vectors of income. To generate counterfactual incomes, Equation (3.14) is subtracted
from the respective total income for every household. This is done in order to remove the
very short run effects of the policy on incomes — the instantaneous income effects. On the
other hand, a second order approximation of income changes due to price supports is
computed by adding the second and third parts of Equation (3.15) on household income.
Next I will discuss the computation of Equation (3.14) and the second and third parts of
Equation (3.15).

Equation (3. 14) is the product of the elasticity and the percentage change in price
due to the policy. As in Equation 3.12, the elasticity is computed as value of maize
production less consumption value divided by total household income. Maize output, sold
quantity, and prices are available from the panel dataset and therefore computation of
production value and sales value is relatively straightforward. To compute consumption
value, I ﬁrst subtracted sales value from production value to get retained value. The value
of production carried over in the next production year was then subtracted from the
retained value to get the value of production consumed at home. I then added the value of
maize purchases for home consumption to the value of production consumed at home to
get maize consumption value. The last step was to subtract maize consumption value
from production value and dividing the result by total household income to get the maize
price elasticity of income. Once I obtained the production and consumption value, the
second part of Equation (3.15) was easy to do. I just multiplied the production and
consumption values with zonal supply and demand elasticity in table 3.1. Computation of

the third part posed some problems since income from labor hired out on maize farms

35

could not be obtained directly from the data. What was available in the data was income
from labor hired out on all farm enterprises. A simple approach was used to apportion
these proceeds to maize. We multiplied total proceeds by the regional ratio of gross value
of maize output to total farm output. Payments to hired labor on maize ﬁelds were
directly obtained from the data. 1

The computations narrated above were done for three different changes in the
price of maize - 15%, 19.7%, and 25%, resulting to 3 pairs of the 2 income vectors. The
19.7% impact emanates from the analysis in Jayne et al (2005a), and the other two are
simulations within a 5% band. These simulations are used to simultaneously reﬂect
possible variations in the magnitude of price changes across regions and to capture any
sensitivity effects on poverty to the magnitude used. Stochastic (poverty) dominance tests
were then conducted separately for each of the 3 pairs of counterfactual incomes and
incomes with effects of price supports. The next section presents the results of poverty

dominance tests.

3.4 Results

Commonly used poverty lines for Kenya include the World Bank $1 a day per
person poverty line (approximately $30 per month per person), and the Welfare
Monitoring Survey (WMS) Kshs 1,238 per adult equivalent per month poverty line4,
which is approximately $16 per month per adult equivalent (GOK 2000). The wide

difference between these poverty thresholds poses a potential source of uncertainty when

 

‘ This poverty line is for 2000 Kshs and therefore it does not need adjustments since the pooled incomes are
also in 2000 Khs

36

welfare rankings are based on one and not the other - the identiﬁcation problem. To
avoid such, ﬁrst degree poverty orderings are proclaimed only when they hold for the
World Bank threshold. This is because the World Bank poverty line (approximately $30
per month per person) nests the WMS threshold ($16 per month per adult equivalent) and
all other poverty lines less than $30 per person per month. We consider second degree

dominance tests whenever ﬁrst degree dominance tests are inconclusive.

Figures 3.1, 3.2, and 3.4 provide ﬁrst degree poverty dominance curves for 19.7%
increase in the price of maize for coastal lowlands, eastern lowlands, and western
lowlands respectively. Simulated dominance curves for 15% and 25% increases are very
similar to the ones for 19.7 % in all zones, and therefore are not presented here. However,
their corresponding crossing points are provided together with the crossing points for

19.7% in table 3.3

Figure 3.1. First Order Poverty Dominance Curve for Coastal Lowlands

Coastal Lowlands: F Irst Order Poverty Dominance Curves (1 9.7% Increase In price of malze)

«of . ..... - p

0.9+ ------ '
0.8 a

 

 

0.7?

Headcount ratio
0
g b.

9
a

0.3»?
02-}

0.1‘.‘

D
)

 

J A A L A

o_4:2:::::.::::¢:;..:::....
O 5 1O 15 2025303604650566005 7O 7500050005100105110115120125130
MorthlypereduequvelerthcuneatlJOUSS)
[— Courtertectuel income -------- income with ettects oi price swpodsj

 

 

37

Figure 3.2. First Order Poverty Dominance Curve for Eastern Lowlands

Eastern Lowlands: First Order Poverty Dominance Cums (1 9.7% increase in price of maize)

 

1.0"
0.9 i _, ---------
0.8l*

0.74? .'

P
a

Headcount ratio
0
*9

0.1" .'

j A J A A A

 

 

 

30:5 40
MontHypereduAequlveierthcomeOMUSS)

[— Courieriectuel income -------- hcome wih etiects of price supports J

o. :::::
0610152026

Figure 3.3. First Order Poverty Dominance Curve for Western Lowlands

46 50 55 60 05 70 75 80 85 W 05 100105110115120125130

Western Lowlands: First Order Poverty Dominance Curves (1 9.7% increase in price oimalze)

 

 

1.0it ..

 

0.9 »
0.3 i

P
2'
\

P
a

P
3’

0.4" "

Headcount ratio

02»;

0.11

v "I--.

 

 

 

«‘0 so ab we so oh 100
WpermmvelertmaneQWOUSS)
[—Comtertectuelincane -------- hcomewith eiiects of price supports 1

A A
I Y

0 1O 20 30 110 120

38

130

143

150

Table 3.3. First Order Poverty Dominance of Counterfactual Incomes over Incomes with
Effects of Price Controls (Headcount ratio higher with Price Controls): All lowlands

 

 

 

 

 

 

 

 

 

Broad First Crossing Point of Cumulative Income Distn’butions Robustness to the
Agro- (Upper Bound Poverty Line for Headcount ratio) World Bank and
climatic US Dollar (8) amount Amount in Kenya shillings (Kshs) the WMS poverty
Zone (std deviation) (std deviation) line

15% 19.7% 25% 15% 19.7% 25%
Coastal 57.49 57.46 57.30 4,486 ' 4,484 4,471 Satisﬁes both
Lowlands (0.00) (0.00) (0.00) (0.00) (0.00) (0.00)
Eastern 69.90 80.43 80.43 5,455 6,276 6,276 Satisﬁes both
Lowlands (12.53) (8.64) (8.80) (978) (674) (687)
Western 49.99 75.44 74.68 3,901 5,887 5,828 Satisﬁes both
Lowlands (0.00) (5.58) (4.1 1) (0.00) (435) (321)

 

 

 

 

 

 

 

Referring to ﬁgures 3.1 through 3.3, the ﬁrst point at which the two cumulative
income distributions cross is the upper-bound poverty line for the headcount ratio. For
any level of income below the crossing point, the value of the cumulative distribution of
counterfactual incomes is lower than that of incomes with effects of price supports.
Because each of the two distributions has a single income observation from every
household, the distribution with a higher cumulative value has more observations with
values less than the crossing point, and hence a higher probability for lower incomes. It
therefore follows that if we take the crossing point or any level of income less than the

crossing point to be a poverty line, the headcount ratio will be higher with price supports.

The headcount ratio is estimated at the vertical axis and incomes (variable poverty
lines) are in the horizontal axis. For example, in ﬁgure 3.1 and ﬁrst row of table 3.3, the
two cumulative distributions cross at US$ 57.46; which is higher than the World Bank
(US $30) poverty line. Further, the cumulative value of the distribution of counterfactual
incomes is below that of incomes with effects of price supports at any value less than the

crossing point. This means that if any income level below the crossing point is taken as a

39

yardstick (e.g. the US $30 World Bank poverty line), the distribution of income with
price supports has a higher probability for lower values compared to counterfactual
distribution of income with no price supports. A higher probability for lower values
translates to a higher headcount ratio, as measured in the vertical axis. We conclude that
NCPB operations in the maize market have increased the number of the poor in Coastal
Lowlands. Figure 3.2 and 3.3 show similar results for Eastern Lowlands and Western
Lowlands respectively. Corresponding crossing points for ﬁgures 3.2 and 3.3 are
respectively summarized in the second and third rows of table 3.3. The lowlands are
characterized by poor soils, poor rainfall, and climatic conditions that are not conducive
for high yielding maize varieties. CBS (2003) shows that some of the major poverty spots
in the country are found in coastal lowlands and western lowlands. Evidently, an increase

in the price of maize exacerbates poverty in these regions.

In contrast to the lowlands, the Kenyan highlands (western highlands and central
highlands) boast high agricultural potential and favorable rainfall patterns. Crop
production is highly diversiﬁed and well serviced by a decent physical infrastructure.
Other than maize these regions are well suited for coffee, tea, horticulture, and dairy
enterprises. Coffee, tea, and horticulture are major income earners and maize is largely
grown as a food crop. Figures 3.4 and 3.5 show results similar to those obtaining in the
lowlands. Corresponding crossing points are provided in table 3.4. We conclude that the
policy has lead to an increase in the incidence of poverty in Western Highlands and

Central Highlands.

4O

Table 3.4. First Order Poverty Dominance of Counterfactual Incomes over Incomes with
Effects of Price Controls (Headcount ratio higher with Price Controls): All Highlands

 

 

 

 

 

 

 

 

 

 

 

 

 

Broad First Crossing Point of Cumulative Income Distributions Robustness to the
Agro- (Upper Bound Poverty Line for Headcount ratio) World Bank and
climatic US Dollar (3) amount Amount in Kenya shillings (Kshs) the WMS poverty
Zone (std deviation) (std deviation) line
15% 19.7% 25% 15% 19.7% 25%

Western 43.01 47.21 56.89 3,356 . 3,684 4,439 Satisﬁes both
Highlands (3.83) (1 1.94) (9.08) (299) (932) (709)

Central 92.36 142.90 123.68 7,207 11,151 9,652 Satisﬁes both
Mghlands (40.89) (22.94) (47.55) (3,190) (1,790) (3,71 1)

 

Figure 3.4. First Order Poverty Dominance Curve for Western Highlands

Western Highlands: First Order Poverty Dominance Curves (1 9.7% increase in price of maize)

 

1£P

0.9 l
0.8 1*

03%

.°
as

01?

O!»

Headcount ratio

0.3-
02» ,5

0.1».-

 

 

 

J

 

A

 

102030405060708000100110
MoulyypereduleiuvelernmneOMUSS)

 

41

120 130 140 160

Courtertectuel income ------- hcome with effects at price supports J

 

Figure 3.5. First Order Poverty Dominance Curve for Central Highlands

Central HIOMOIIGSZ FII‘SI Order Poverty Dominance Curves (1 9.7‘ IIICI’OOSO Ill price OII'IIOIZO)

 

1.0'

 

0.9 ..
0.8 - """"""

0.7"
0.6i
0.5" .

0.4"

Headcount ratio

0.3'r
0.20

0.1" I:

 

 

J A

 

o" 4 4. : s c 4. c c c t C : c . . ‘r : : :
o10203040soeo7oeoco100110120130140150100170190100200
MadlyyperedullemlvalerthcomeatIIlUSS)

[—Caurteriectuel 'ncome -------- hcomewlheiiects of price supports ]

 

In the western transitional zone, the crossing point of ﬁrst degree stochastic
(poverty) dominance curves does not nest the World Bank threshold (ﬁgure 3.6 and table
3.5). Therefore poverty orderings in this region will be based on the poverty gap measure,
which leads us to consider second degree dominance test (Figure 3.7). This test
determines the cumulative difference of the area under the distribution of incomes with
NCPB operations from the area under the distribution of counterfactual incomes. The
latter dominates the former as long as the cumulative differences remain non-positive.
Figure 3.7 shows that the differences are non-positive across the entire range of income.
This means that the distribution of counterfactual income without price supports
dominates in the second degree the distribution of income with price supports.

Equivalently, we say that if any level of income is taken to be a poverty line then income

42

shortfalls from any such poverty line are higher with price supports. We therefore
conclude that in this region, the NCPB maize operations reduce incomes among those

already poor, but do not appreciably affect the numbers of the poor.
Figure 3.6. First Order Poverty Dominance Curve for Western Transitional

Western Transitional: F lrst Order Poverty Dominance Curves (1 9.7% increase in price of maize)

 

 

1. . J“
0.9 r "r ..........
0.8 ' .-"'

0.7"
0.6 '

0.6"

0.4‘

Headcount ratio

0.3-
02»

01:."

 

A

 

 

1'omm1meb7ommimm1mieoiioieo
WperedileqlvaiertmomemLiSS)
[— Canterlectudlncome -------- hcomewithetlects otprlcesupports]

0.

43

Figure 3.7. Second Order Poverty Dominance Curve for Western Transitional

Second Order Poverty Dominance: Western Transitional (1 9.7% increase in price of maize)

 

O.

-0 .251i

55
S

0.75“

“LCD"

.125»

Cumulative difference

4.50"

-‘l .75"

4

J A

A

 

A A J

 

 

 

[—Cmawvedﬂerencewve ]

 

0.00 0.05 0.10 0.35 020 055 0.50 0.55 0.00 0.45 050 0.55 0.50 0.05 0.70 0.75 0.00 005 0.00 095 100
Peroerlles (p)

Table 3.5. First Order Poverty Dominance of Counterfactual Incomes over Incomes with
Effects of Price Controls (Headcount ratio higher with Price Controls): Western

 

 

 

 

transitional
Broad Agro- First Crossing Point of Cumulative Income Distributions Robustness to
climatic Zone (Upper Bound Poverty Line for Headcount rang the World Bank
US Dollar (3) amount Amount in Kenya shillings and the WMS
(std deviation) (Kshs) poverty line
(std deviation)
15% 19.7% 25% 15% 19.7% 25%
Western 29.52 30.10 30.61 2,304 2,349 2,389 Satisﬁes only
Transitional (3.96) (3.48) (3.18) (309) (272) (248) WMS poverty
line

 

 

 

 

 

 

 

 

 

 

Conditions in the western transitional zone with regard to land potential, market

infrastructure, and crop mix approach those in the high potential maize zone (HPMZ). In

Kenya, a large proportion of marketed maize is grown in the high potential maize zone.

The region boasts excellent conditions for growing maize and wheat, and farm sizes are

relatively large. Among the seven agro-climatic zones considered, it is only in this zone

44

where the price-increasing maize policy does not increase poverty as measured by the
headcount ratio and the poverty gap measure. In Figure 3.8 and table 3.6, the crossing
point of the two cumulative distributions is below both the World Bank poverty line and
the WMS threshold. Similarly, Figure 3.9 and Table 3.7 for second degree dominance
test shows that cumulative differences in the area under the two distributions remain non-
positive only up to the 45m percentile of income, which is equivalent to $27.17 or Kshs

2,120. This threshold is lower than the World Bank poverty line and therefore the test is

inconclusive.

Figure 3.8. First Order Poverty Dominance Curve for High Potential Maize Zone

Hill] Potential Zone: FII'SI OI'IIOI' DOMIIIOIICB Curves (1 9.7% "1079880 "I maize DI'ICO)

 

 

 

 

1.0+ "m“: ,,,,, .21 ---.

-"---
0.9 7' 4”
ﬂ

9 '3
a .
\

P
as

0.4"

Headcount ratio
0
cr

0.3-
02. ,-

0.1"):

 

 

A_ A

 

0. . c € € 4. ¢ 5 i 3 4 i 3 c . v t

O 25 5O 75 100 125 150 175 200 225 250 275 300 325 350 375 430
Morﬂvypermeqlvalerthcomeamo US$)

(— Comterfactuel income -------- hcome wth effects of price supports ]

45

Figure 3.9. Second Order Poverty Dominance Curve for High Potential Maize Zone

Second Order Poverty Dominance: High Potential Ilaize Zone (1 9.7% increase in price of maize)

 

10'

 

Cumulative difference

 

A

 

L

 

 

[—CumtdivedtferencecuveJ

 

0.00 0.05 0'10 0.35 020 025 0.00 0.55 0.30 0'45 0.50 0.55 0.00 0.05 0.70 0.75 030 005 000 005 1.00
Percertles (p)

Table 3.6. First Order Poverty Dominance of Counterfactual Incomes over Incomes with
Effects of Price Controls (Headcount ratio higher with Price Controls): High potential

 

 

 

 

maize zone
Broad Agro- First Crossing Point of Cumulative Income Distributions Robustness
climatic (Upper Bound Pov Line for Headcount ratio) to the World
Zone US Dollar (S) amount Amount in Kenya shillings (Kshs) Bank and the

(std deviation) (std deviation WMS
15% 19.7% 25% 15% 19.7% 25% Poverty line

High 13.59 14.49 14.49 1,061 1,131 1,131 Satisﬁes
Potential (2.38) (1.83) (1.56) (186) (143) (122) None
Maize Zone

 

 

 

 

 

 

 

 

 

 

Table 3.7. Second Order Poverty Dominance of Counterfactual Incomes over Incomes with
Effects of Price Controls (Poverty Gaps Higher with Price Controls): High potential maize

 

 

 

 

zone
Broad Agro- Amount at which cumulative difference between income distributions Robustness
climatic reaches zero (upper bound poverty line for poverty gap measure) to the World
Zone US Dollar (S) amount Amount in Kenya shillings (Kshs) Bank and the
(std deviation) (std deviation) WMS
15% 19.7% 25% 15% 19.7% 25% povenyline
High 24.39 27.17 30.01 1,903 2,120 2,342 Satisﬁes only
Potential (1.89) (1.82) (2.07) (147) (142) (162) WMS
Maize Zone poverty line

 

 

 

 

 

 

 

 

 

 

46

In summary, we ﬁnd that the effects of the price-increasing policy of the NCPB is
largely inﬂuenced by the proportion of net purchasing and net selling rural households in
each zone. As shown in Table 3, the marketed maize output in Kenya is concentrated in
one zone (High-Potential Maize Zone). Most other rural areas of Kenya derive the bulk

of their cash income from other crops, non-farm income, and livestock.

3.4 Conclusions

This chapter serves two purposes. The ﬁrst is to reveal the conceptual framework
used in measuring the effects of price policy on welfare, and the second purpose is to
examine the impact of the price policy on poverty. Methodologically, this study extends
the previous disequilibrium model used in Deaton (1989), Budd (1993), and (Barrett and
Dorosh 1996) and provides a second order approximation of the equilibrium income

changes.

Second round effects on incomes are estimated with an assumption of equal
supply and demand elasticity for maize farmers across the income divide. It can be
expected that factor ﬁxity would be different between the poor and the rich, and therefore
the supply response could be higher for wealthy farmers who have more factors at their
disposal. On the other hand, since maize purchases comprise a larger budget share among
the poor, the ensuing augmented tilt in the budget constraint of the poor will probably be
met by a demand response larger than that of rich farmers. To examine if possible
differences could affect the results, I conducted a sensitivity analysis using different
demand and supply elasticity for 3 income terciles. For the lower income tercile, the

supply elasticity in table 3.1 was reduced by 20% while the demand elasticity was

47

increased by the same margin. Similarly, I reduced the demand elasticity by 20% for the
higher tercile and increased the supply elasticity by the same margin. The demand and
supply elasticity in table 3.1 were maintained for the middle tercile. These changes did

not affect the results.

Usually, welfare evaluations based on the commonly used F GT poverty measures
employ a single poverty line. As such, it is possible that results could be reversed when a
different poverty line is used; an ambiguity that is clearly undesirable. This paper does
not suffer the weaknesses posed by the identiﬁcation problem. The analysis has exploited
the parallel between stochastic dominance and FGT measures of poverty to generate
poverty rankings that hold for a wide array of poverty lines, including the commonly
used World Bank $1 a day poverty line and a threshold used by the government of
Kenya. Further, the analysis is conducted for all price increases within 5% standard
deviation of the impact suggested by a recent study on the effects of the policy on price

levels.

Results show that the joint effects of NCPB price supports and tariff on imported
maize increase the number of poor farmers across Kenya, except in the maize surplus
regions of the Rift Valley and Western provinces (HPMZ and Western Transitional
zones). In the Western Transitional zone the policy seems to reduce incomes among those
already poor, but does not appreciably affect their numbers. Looking at the varied
poverty effects of the policy in different regions, we infer that empirical work at the
national level may not be appropriate as it would mask important regional differences,
especially in situations where farming conditions are not homogenous. Differences in

agro-climatic conditions, land potential, crop diversiﬁcation, and household incomes

48

proﬁle interact to produce varying intensities of the impact of changing food prices on

poverty.

49

CHAPTER 4
MAIZE PRICE POLICY AND INCOME DISTRIBUTION
Abstract

Most empirical studies on the subject of welfare effects of changing food prices are
focused on the income distribution aspect. The empirical methods used have barely
evolved beyond the non-parametric framework used in the ﬁrst analysis. This paper
considers a partly linear semi-parametric model to control for the inﬂuence of variables
that not only inﬂuence the elasticity but which are also correlated with income. The
results indicate that households at the lower points of the income range loose at least 25%
of their incomes while the wealthiest either gain some proportion of their incomes or
remain unaffected. The gains improve as initial incomes rise and the losses become more
adverse as initial incomes fall. This general pattern is common in all the regions

considered.
4.1 Introduction

Empirical studies on the welfare effects of food pricing policies have mainly
addressed the question of how price changes affect the distribution of income. Most
studies on this subject have largely employed non-parametric methods following their
ﬁrst use in Deaton (1989). Deaton (1989) considered the distribution of ‘net beneﬁt
ratios’ or ‘net consumption ratio’ conditional on expenditure to analyze the income

distribution effects of price changes. The key ﬁnding from the study was that rice price

 

5 Net beneﬁt ratio (NBR) also called Net consumption ratio is calculated as production value less
consumption value as a ratio of expenditure

50

increases in Thailand had little ﬁrst order impacts on the welfare of the poor and the rich.
It is farmers in the middle of the income distribution that were affected most either as
gainers or losers. Budd (1993) found that changes in welfare due to changing prices of a
single commodity or a group of commodities were quite small in Cote d’I Voire. This
was the case because net purchases of individual commodities or a group of them formed
a small portion of expenditures (less than 5%). This pattern was found to be stable across

welfare levels.

Barrett and Dorosh (1996) focused on rice in Madagascar. They examined
marketable surplus and net rice sales conditional on per capita income and per capita
land, an approach which assumes that the proportion of production consumed at home
consumption is inelastic to changing prices. The results they obtain are different from
Deaton (1989) in that both marketable surplus and net rice sales were lowest among the
income poor and land poor, but highest among the rich and those highly endowed with
land assets. They conclude that increasing rice prices would mostly hurt the poor and that

the gains would gradually improve as income and land endowment increases.

The non-parametric regressions used in all the previous studies were based on
arbitrary bandwidth parameters - usually over-smoothed, and therefore the convergence
properties of the estimated conditional expectation functions are highly compromised.
Over-smoothing of regression curves can cause very misleading conclusions. For
example, if the regression curve based on the optimal bandwidth parameter (the
consistent estimate) is generally upward sloping but with a dip in the middle, over-
smoothing would ﬂatten the dip and give a gentle upward slopping curve. The

information provided by the clip will clearly be lost, and therefore conclusions based on

51

an over-smoothed regression curve could be very misleading. This critical shortcoming in
previous analyses was probably entertained because of the complexity involved in
computing optimal bandwidths. However, recent developments in statistical software
have made such computations inexpensive. In this study, all regression curves are
estimated using optimal bandwidths computed using the cross-validation method. The

DAD software is used for this purpose.

This study provides two methodological extensions to earlier work. Using a
second order approximation of equilibrium income changes, we are able to account for
the effects of the policy on supply response, demand for maize, and the rural labor
market. This is in contrast to previous studies which employed a disequilibrium ﬁrst
order analysis which inevitably gives estimates that are at the lower bound. The second
extension regards the use of a partly linear semi-parametric model which enables us to
control for the inﬂuence of household demographics, land access, input and output prices,
rainfall, physical infrastructure, and household characteristics such as levels of education.
The model is estimated using the Generalized Speckman Estimator (Speckman 1988).

Results suggest that household size (adult equivalent), age of household head, and
the incidence of female-headed households are negatively related to gains from the
policy. As it could be expected, the level of adult education and the value of farm assets
have a positive relationship with gains from the policy. The income distribution bias of
increased maize prices is not very different across the regions. Households with incomes
at the lowest points of the range lose at least 25% of their incomes while the wealthiest
either gain some proportion of their incomes or remain unaffected. These patterns are

very similar to those observed with regard to rice in Madagascar (Barrett and Dorosh

52

1996). We therefore conclude that the policy engenders inequality in rural Kenya on top
of the adverse consequences on poverty that were found in chapter 3.

The chapter is organized as follows; the next section provides a conceptual
framework for estimating distributional effects of the policy, a discussion that is centered
on the speciﬁcation and estimation strategy of the partly linear semi-parametric model.

Results are presented in section 3 followed by conclusions in section 4.

4.2 Conceptual Framework

The poverty orderings in chapter 3 summarize the effects of the policy on
incomes of households that are already poor and those who become poor as a result of the
policy, but they ignore potential effects on incomes of the non poor who remain non
poor. Questions such as whether rich farmers beneﬁt more relative to poor farmers, or
vice versa, and how that differs from one region to another are important for policy
makers. The next section presents a model to study the effects of the policy on the

distribution of income.
4.2.1 The Semi-parametric Model

The magnitude of changes in income as a result of the policy are inﬂuenced by
factors determining production, marketing, and consumption levels; such as land assets,
demographic characteristics, input and output prices, market access, and household

6

characteristics e.g. levels of education . Some of these factors are also correlated with

income; for example, levels of education could move together with total household

 

6 Table 3.3 summarizes variables considered in the model

53

income through their effects on non-farm income and salaried income. Others such as
market access, land assets, and input and output prices determine the degree to which
adjustments are made in markets for other commodities. The semi-parametric model
controls for the inﬂuence of these factors in estimating the expected changes in income at
different levels of it. These factors are assumed to have a parsimonious linear relationship
with the dependent variable hence the parametric part of the model. Expected changes in

income are estimated non-parametrically.

The panel data representation of a partly linear model takes the form
y. = z..6+m(x.)+c. +u... (4.1)

Where, x” is income of household i at time t and 2,, is a vector of variables that include

land assets, demographic characteristics, and household characteristics, among others.

The dependent variable y" is the second order approximated equilibrium changes in
income. c, is a time constant unobservable and u" is a zero mean disturbance term.

There are two reasons why semi-parametric estimation is preferred. The main thrust is
that it provides a versatile method of exploring the relationship with respect to income
without reference to a parametric speciﬁcation, more so because economic theory does
not offer much guidance on functional forms. As such the approach is well suited to
analyze the effects on the distribution of income because the expected changes in income
can be estimated at every point along the income continuum. The second advantage
relates to the fact that income (an explanatory variable) and the second order
approximated changes in income (the dependent variable) are jointly determined, which

makes parametric estimation cumbersome.

54

Unlike in previous studies, we use Optimal smoothing parameters in all non-
parametric regressions to insure consistency of the conditional expectation function. The
problems of using arbitrary smoothing (bandwidth) parameters in non-parametric
regressions are similar to those of functional form misspeciﬁcation in parametric parlance
Hardle (1990). Regression curves based on arbitrary bandwidth parameters are usually
over-smoothed and lead to imprecise estimation of the conditional expectation function.
For example, if the regression curve based on optimal bandwidth parameter is generally
upward sloping, but with a dip in the middle, an over-smoothed curve will ﬂatten the dip
leading to a gentle upward sloping curve. The resulting gentle upward sloping curve is a
false conditional expectation function and therefore statements based on it could be very

wrong.
4.2.2 The Estimation Strategy

A variety of techniques have been used to estimate partly linear models. Robinson

(1988) constructed a feasible least square estimate of ,8 based on separate estimates of
the non-parametric component by a Nadaraya-Watson estimator. Under some regularity

conditions, ordinary least squares (OLS) regression of y,, on 2,, alone consistently and
efﬁciently estimates ,8. Required for consistency is E(z,,m(x,,))=0, a condition
attainable when either E (2,,): 0 , or through statistical independence between x,, and

2,, . This approach is not considered here because there is evidence in the literature that

income is correlated with variables such as asset values, land access, and some household

characteristics.

55

Other techniques that could be used include Spline smoothing. This approach has
been used in a variety of studies such as Green and Silverman (1994); Engle, Granger,
Rice and Weiss (1986); Heckman (1986); and Rice (1986) to derive penalized estimators

of ,6 and m. In this context, a penalty parameter is used to reﬂect the degree of
compromise between minimizing the residual sum of squares (in estimation of ,6 ), and

smoothing of m (in estimation of the regression curve). Rice (1986) and Schimek (1997)

show that the resulting estimators of ﬂ and m are inconsistent.

This study uses the Generalized Speckman Estimator (Speckman 1988) because

the properties of the estimators are well known, and also because the procedure is

relatively easy to apply. Speciﬁcally, ,5 achievesJ; consistency while the rate of

convergence of the estimator of m(x,,) and its derivatives is lower than J; (Hardle

1990). The procedure is essentially an application of the Frisch-Waugh Theorem,
formally presented and stated in Greene (2003) thus “In the linear regression of vector Y

on two sets of variables, X, and X 2 , the subvector ,8, is the set of coefﬁcients obtained
when the residuals from a regression of Y on X, alone are regressed on the set of
residuals obtained when each column of X 2 is regressed on X ,. To demonstrate the

procedure, I begin by taking expectations conditional on x,, from Equation (4.1).

150.. I x. )= E(z..ﬂ I x. )+ E {m(x. )I x. }+ E(c. I x. )+ E(v. I x. )- (4.2)
Subtracting the above from Equation (4.2) gives

yl! _E(yu Ix”): {zll -E(Z,, IX,,)},B+C,- —E(cl lel)+Ull _E(Ull lxu)‘ (4'3)

56

Since E {m(x,,)l x,,}=m(x,,). The composite disturbance term in Equation 4.3

u —E(u,, |x,,) has a mean of zero since by deﬁnition Equation (4.1) implies

E(u,, |x,,,2,,)=0. It is important to note that y,, -E(y,, |x,,) and {2,,—E(z,, |x,,)}
respectively give expected values of the dependent variable and the covariates after

removing the effects of income.

In order to use Equation (4.3) to estimate ,6, we will ﬁrst replace the terms
E(y,, |x,,) and E(z,, |x,,) by their non-parametric estimators. Sample analogs of
E (y,, |x,,) and E (2,, |x,,) are obtained using the Nadaraya-Watson estimator - in the

same way that the non-parametric part of model is estimated. The discussion on non-
parametric regressions is reserved for section 4.2.2.2. However, it is worth noting that it
is the residuals from Equation 4.3 that will be used to estimate the non-parametric part of

the model. These residuals represent the part of variation in y,, that is not explained by

the covariates and which therefore could be attributed to variation in household incomes.

The next section will discuss various approaches that could be used to estimate

,6 . The covariates in the parametric part of the model include various household

characteristics such as age of the head of household, highest education level by an adult
member, female head dummy, and household size (measured by adult equivalents),
among others. District dummies are used to account for differences in maize prices and
prices of production inputs such as fertilizers and hybrid seeds. Variables such as distance
to motorable road and distance to hybrid seed seller are used to capture market access and
the level of infrastructure development. Weather variations are captured by rainfall levels

in the main planting season.

57

4.2.2.1 Estimating the Parametric Part

A variety of methods are possible for the estimation of )6 from Equation 4.3. First

I considered pooled OLS. Standard tests on results from pooled OLS suggest the
existence of both heteroskedasticity and serial correlation. The latter is usually related to
a time-constant unobservable factor (unobserved heterogeneity) in the error term
(Wooldridge 2001). Correlation between the unobserved time-constant factor and any of
the explanatory variables renders pooled OLS inconsistent and therefore I considered
other methods such as ﬁrst differencing and ﬁxed effects estimation. The motivation for
ﬁrst differencing is that the time constant unobservable variable is removed in the
differencing process. However, the process would also remove other time constant
variables such as education, land size, and demographic characteristics, and therefore
partial effects of these variables cannot be estimated even when they are constant only for
a section of the sample.

The ﬁxed effects method is particularly attractive because it allows for arbitrary
correlation between the unobservable factor and any of the observable explanatory
variables. However, the method suffers the same weaknesses as ﬁrst differencing — the
effect of time constant observables cannot be estimated. Because of the nature of the
estimator, it is important that the effects of all variables in the parametric part of the
model are removed before estimating the non-parametric part. This leads us to the
random effects estimator. With this method it is possible to get precise estimates of the
partial effects of time constant observables such as education, household characteristics,
and demographic characteristics. The key assumption that would allow us to proceed this

way is that the time constant factor is uncorrelated with any of the explanatory variables.

58

Random effects estimation gives inconsistent estimates whenever the assumption is
violated.

In the context of this work, random effects approach becomes attractive and
practical since the interest is on the partial effects of variables some of which do not
change over time. Before proceeding with the random effects procedure, I conducted the
Hausman test to look for systematic differences between the random effects estimates and
coefﬁcients estimated through ﬁxed effects. The null hypothesis in the Hausman test is
that random effects estimator is consistent and efﬁcient. Results show that we cannot
reject the null that random effects estimator is consistent and efﬁcient for ﬁve out of the
seven regions. In the two regions where the null was rejected, the ﬁxed effects procedure

suggested that none of the variables in 2,, had explanatory power. This means that all the
variables in 2,, could be ignored from the models for those two regions.

Estimation of ,6 is largely an intermediate process which is used to remove the
part of changes in income that is explained by the 2,, variables. It is the residuals from

this regression that will be used in the next step to obtain non-parametric estimates of the
expected changes in income at its different levels (as will be discussed in the next

section). Residuals from the ﬁxed effects estimator will still contain that part of 2,, that is

time invariant e.g. the size of land owned and education levels. On the other hand, if we

use random effects then we would fully remove all the variables in 2,, and the

accompanying residuals will just contain that part of income changes which is not

explained by 2,,. The next section will discuss the non-parametric estimator used for the

regression between the residuals and log of income.

59

4.2.2.2 Non-parametric Regressions

After obtaining the residuals from Equation (4.3), the last step of the Generalized

Speckman Estimator is to estimate m(x) using the Nadaraya-Watson estimator from

E(yll —leﬂlxll): m(xll)+ull (4'4)

In the remaining sections we use the standard notation for the explanatory variable y,, in
place of (y,, — 2,,,6 | x,, ). With this change in notation we can deﬁne the conditional
expectation m(x,,) as

m(x.>= 150. ix. =x)= ly. [17%.{3}, (4.5)

Where, the numerator f” (x, y) is the joint density of x,, and y,,, and the denominator
f, (x) is the marginal density of x,,. The conditional expectation in Equation (4.5)

deﬁnes the regression curve of y,, on x,,. Following Hardle (1990) 751,, (x,,) the pooled

estimator of the conditional moment is given as

 

7' "' . (4.6)

Equation (4.6) is called the Nadaraya—Watson estimator. Where, n is the sample size, T

is the length of the panel, and x,, and y,, are pair-wise values of the log of income and the

60

changes in income after removing the effects of 2,,. The estimator comprises a weighting

function for each y,, deﬁned by

w,,(x,x,,)= n Ifh(x-xu) . (47)

ZZK,(x-x,,)

is] (=1

 

The shape of the weight function is determined by the kernel density function K,
whereas the bandwidth parameter h determines the size of the weights that sum to unity.

In this study K h is the Gaussian Kernel which takes the form

 

exp[— (ii—LL] . (4.8)

Unlike most kernel functions, the Gaussian kernel is not bounded on x,, which means

that in theory all data points will be brought into every estimate. However, points outside

three standard deviations hardly make a difference in practice.

Under ‘independent and identically distributed’ (iid) assumptions on u,,, the
empirical regression curve riih(x,,) is consistent for the conditional expectation m(x,,).

The rate at which 751,, (x,,) converges to m(x,,) is determined by the Kernel (K), and the

bandwidth (h ). Hardle (1990) shows that of the two determinants, the bandwidth is the

most critical in the precision of ﬁt, (x,, ). Bandwidth selection methods include leave-one-

out cross validation, use of penalizing functions, and the plug-in method. A discussion of
these methods and the optimality of resultant bandwidths can be obtained from Hardle

(1990). Hardle and Marron (1985) have shown that the leave-one-out cross validation

61

method leads to optimal bandwidth selection. The bandwidth parameter selected using
this method minimizes the various asymptotic equivalent global measures of the distance7

between 771,, (x,,) and m(x,, ). The estimator n‘r, (x,,) is said to be ‘asymptotically optimal’

when the cross validation optimal bandwidth parameter is used, because it achieves the

optimal rate of convergence to m(x,,) (Hardle 1990).

The iid assumption on u,, is not tenable when panel data is used. This is because

the error term in one period might be correlated with that of another period for the same
household. The correlation speculated here may be related to unobserved heterogeneity.
Ignoring correlation in errors leads to a choice of a small bandwidth parameter and

compromises the convergence properties of iii, (x,, ); the reason being that the regression

method will interpret the existing correlation in the errors as part of the regression curve.

Given a two year panel with ( y,, ,x,,) ordered chronologically and stacked on top of one

another, allowing for ﬁrst order serial correlation sufﬁces for the computation of optimal
17. Usually regressions that allow for serial correlation in the error term lead to a higher
bandwidth parameter than those that assume independence of the error term (Hardle

1990).

4.3 Data Construction

The data used in this analysis is compiled from the Tegemeo household surveys
described in section 2.2. Household and demographic characteristics variables are easy to
compute directly from the demography modules in 2000 and 2004 survey data. Similarly,

the panel data has readily available information on land access, asset endowments, and

 

7 Global measures evaluate convergence at all points-the mean square error looks at pointwise convergence.

62

infrastructure variables. Tegemeo also provided data on rainfall which they had originally
sourced from the department of meteorology in Kenya.

Tegemeo data has very detailed information on various sources of income which
include; crop income, livestock income, salaries and remittances, and incomes from
business and informal activities. Incomes from each of these sources are computed and
then summed up to give total household income. These computations are done separately
on the 2000 and 2004 survey data. The vector of 2004 total household incomes is then
deﬂated to 2000 using the ofﬁcial inﬂation rate proﬁle in the two years. The data is then
combined to form a two year panel data. Next I used the pooled total household income
vector to generate a vector of counterfactual incomes. Computation of counterfactual
incomes and the second-order approximated percentage changes in income is already

explained in Section 3.3.

4.4 Results

4.4.1 The Inﬂuence of Factors Other Than Income

This section will discuss results from the estimation of Equation 4.3. As discussed
in section 4.2.2.1, the estimation approaches considered in this study are ﬁxed effects and
random effects. With the random effects estimator, the variance due to the time constant
unobservable was found to be higher than that of the idiosyncratic error. This means that
estimates from the random effects generalized least squares are better compared to those
from pooled OLS (Wooldridge 2001). It is for this reason that the random effects
procedure is considered rather than pooled OLS. Table A4.1 through table A4.7 in the

appendix section give results for the seven agro-climatic zones. Only variables that had

63

joint signiﬁcance were retained in the models because the aim of these regressions is to
get residuals which will be used in the last step of the generalized Speckman estimator.

The Hausman test which was performed for every zone suggests that the random
effects estimator is consistent and efficient in ﬁve out of the seven zones. These ﬁve
zones are coastal lowlands, eastern lowlands, western lowlands, high potential maize
zone, and central highlands. The null that the random effects estimator is consistent and
efﬁcient is rejected in the models for western highlands and western transitional. The
coefﬁcients from the two estimators are of the same signs in models for all the regions.

Where signiﬁcant, the inﬂuence of household size is negative. This pattern could
be attributed to prevalence of small farm sizes, to the extent that the marginal labor
productivity of maize is inﬁnitesimal for a majority of households. On the other hand,
maize consumption requirements increase as household size expands. Both of these
forces lead to a reduction in marketable surplus or an aggravated net buyer status. The
age of the head of the household is negatively related to gains from maize price policy
while squared age has a positive inﬂuence which depicts a situation of increasingly
negative welfare effects as age advances. The value of farm assets and the level of adult
education are associated with gains from the maize price policy. Higher levels of
education give households an increased capacity for decision making with regard to
allocation of factors of production to competing enterprises, crop husbandry practices,
and consumption allocations. These results suggest that better decision making capacity
could increase gains to net sellers or mitigate adverse effects to net buyers.

The incidence of female-headed households is associated with losses from

increasing maize prices. Traditionally, women have been excluded from land access

64

mainly because most cultures disallow land inheritance to female offspring’s, yet
inheritance is the most common way of passing the resource from one person to another.
Women starting families outside marriage are mostly trapped into landlessness, which
makes them natural net buyers. When widowed, most women lose control of important
assets such as land to male relatives, even though they still work on the farms. With this
mode of dispossession, households headed by widowed females receive mercy payments
to their labor while controlling males from other households extract rents from all
farming activities. This makes female headed households net buyers just because the
share of production they receive is disproportional to their labor input and consumption

requirements.

4.4.2 Expected Changes in Income along its Range

This section provides results of non-parametric estimates of expected changes in
income at different points along its range. There are two regression curves in ﬁgure 4.1
for the coastal lowlands region, ﬁgure 4.2 (eastern lowlands), ﬁgure 4.3 (western
lowlands), ﬁgure 4.4 (central highlands), and ﬁgure 4.5 for the high potential maize zone.
The ﬁrst one is derived from random effects residuals and the second is obtained from
ﬁxed effects residuals. In the discussion that follows I will be referring to the regression
curve based on the random effects residuals because the Hausman test suggested that the
random effects estimator is consistent and efﬁcient. The regression curve based on the
ﬁxed effects residuals is included so as to demonstrate that those residuals include the

effects of covariates that do not change over time. As a result of this, the regression

65

curves based on the ﬁxed effects residuals are a south-easterly shift of the one based on

random effects residuals in all but the high potential maize zone.

Figure 4.1. Expected Changes in Income in Coastal Lowlands

Coastal Lowlands: Percentage Changes in Income along its Range

 

Proportional income change

 

 

 

Log of Cornerfactual hcome (US$): CVh - 0.291
[ — 2nd Stage Estimator is Random Effects -------- 2nd Stage Estimator is ﬁxed Effects |

The gains or losses from increasing prices are measured by the proportion of
income changes along the vertical axis. In both regions we see that the regression curves
are upward 510ping, which means that households with higher income levels gain at the
expense of those at the lower points of the income distribution. Households at the lowest
point on the income range lose about 50% of their per-capita incomes while those at the
highest end of the range gain by about 30%. There seems to be a threshold of income
after which households turn from being losers to gainers. From ﬁgure 4.1 it can be
inferred that households with per-capita income of about US$135 per month incur zero

proportional changes in income in coastal lowlands. Those with income levels above this

66

threshold experience gains from the maize price increasing policy. The ﬁndings in the
coastal lowlands are very similar to what is observed in eastern lowlands (ﬁgure 4.2

below). A minor difference is that the threshold which separates gainer from losers is a

little bit lower an US$ 11.0.

Figure 4.2. Expected Changes in Income in Eastern Lowlands

EOSIOI’II LOWIOIICSI Percentage Changes Ill IIICOI'IIO along its Range

 

A
A
Y

 

.699
+9

.6
r9

inst-.6
000

 

Proportional income change

 

 

J L A J;

A A
I V

125 1.50 :75 200 2.25 2.50 2.75 3.00 325 350 3.75 4.00 4.25 4.50 4.75 5.00 525
Log of Courterfactual hcome (US$) CVh - 0.260
[ —— 2nd Stage Estimator is Random Effects -------- 2nd Stage Estimator is Fixed Effects J

 

The results from coastal and eastern lowlands are quite different from what we
will see in the western lowlands (ﬁgure 4.3). One of the key differences is that the gains
from the policy are fairly stable after the zero proportional change threshold in western
lowlands, while the regression curves for coastal and eastern were upward sloping
beyond this point. Further, the threshold is much higher at approximately US$ 20.00 per-

capita per-month, which means that households need to be more endowed if they are to

67

gain from the maize price increasing policy. These results are very similar to those from

the central highlands (ﬁgure 4.4).

Figure 4.3. Expected Changes in Income in Western Lowlands

Western Lowlands: Percentage Changes in Income along Its Range

 

 

1 income change

Prpportiona

 

 

.1
' A A L A A A AA A A A A A A J A

0.50 0.75 1.00 125 1.50 1.75 2.00 225 250 2.75 300 325 350 3.75 4.00 4.25 4.50 4.75 500 525
Log of Courterfectuel hcome (US$) CVh - 0.272
| — 2nd Stage Estimator is Random Effects -------- 2nd Stage Est'metor is Fixed EffectsJ

Figure 4.4. Expected Changes in Income in Central Highlands

Central Highlands: Percentage Changes in Income along its Range

 

 

Proportional income change

 

 

, I
~. ’

'0‘!
A AA A A A A A A

 

j A A A A A A

125 1.50 1.75 2.00 2.25 2.50 2.75 3.00 325 3.50 3.75 4.00 425 4.50 4.75 5.00 525 5.50
Log of CorderfectuelhcomeﬂlSSI CV11 -0.203
L—znd Stage Estimdor is Random Effects -------- 2nd Stage Estimator is Fixed Effects J

 

68

Next we will look at the regression curve for the high potential maize zone
(HPMZ). Maize is a major source of cash among farmers in the HPMZ. Other farm
enterprises of choice to farmers in this region include wheat and commercial dairying.
Figure 4.5 suggests that gains from the policy are stable above the threshold income level
in this zone. This ﬁnding is very similar to what was observed in eastern lowlands and
central highlands. However, there are some notable differences; for instance, the
proportion of losses to households in the lowest point on the income range reduces to

25% compared to about 50% in the lowlands and the central highlands.

Figure 4.5. Expected Changes in Income in the High Potential Maize Zone

High Potential Ilalze Zone: Percentage Changes in Income along Its Range

 

 

Proportional income change

0.40"
-0.45"_ ,1‘

 

-O. . . r . . A . r . . . J . . . . A .
0.75 1.00 1.25 1.50 1.75 2.00 225 2.50 2.75 3.00 3.25 3.50 3.75 4.00 4.25 4.50 4.75 5.11) 5.25 5.50
Log of Courterfactuel Income (US$) CVh - 0229

[—20 Stage Estimator isRendom Effects -------- 2nd Stage Estindorisi-‘ixed Effects 1

 

 

The other regions considered in this study are western highlands and the western
transitional zones. As mentioned in the preceding section, the model did not perform well

in these regions. The ﬁxed effects model suggested that none of the variables had

69

explanatory power and the Hausman test led to the rejection of the null that the random
effects estimator is consistent and efficient. Both taken together imply that there is no
need for the parametric part of the model in Equation 4.1. Essentially we could proceed
by just regressing proportionate changes in income on log of per-capita income without
controlling for the effects of other covariates. The ensuing regression curve has a very
similar shape to the one based on random effects. It can therefore be inferred that just like
in other regions of Kenya, the policy hurts households with less income more that it could
hurt those with higher incomes. The regression curves for western highlands and western

transitional are given below in ﬁgure 4.6 and ﬁgure 4.7 respectively.

Figure 4.6. Expected Changes in Income in Western Highlands

WOSIOI‘H Highlands: Percentage Changes III Income along IIS R8090

 

0.14r
0.0"
-0.11

been
{rave

 

I
b?
5.01

\

........

-0.8 i

Proportional income change
3°
\0

e-I u.
' .
'2 9
7 1
~

 

 

A l A

0.75 1.00 125 1.50 1.75 2.00 225 2.50 2.75 300 325 3.50 3.75 400 425 450 4.75 5.00 525
Log of Courterfectud hcome (US$) CVh - 0.224
[—2nd Stage Estimatorls Random Effects -------- Zero Etteots of Covariates ]

 

70

Figure 4.7. Expected Changes in Income in Western Transitional

Western Transitional: POI‘CBI‘IIBOO Changes In Income along its Range

 

 

 

 

 

 

 

0

CD

5

e:

0

0

E

8

OS

'3

t:

02

r:

it

6:

.o,50-- "\ ..........
0.75 100 125 1.50 1.75 2.00 225 2.50 2.75 300 325 350 3.75 4.00 425 450 4.75 500
LogofCourterfactuallncortretlBSICVh-Dm
F— 2nd Stage Estimator is Random Effects -------- Zero Effects of Covariaes j

4.4.2.1 Concerns about Measurement Error

The dependent variable used in the regressions is calculated as the value of net
maize sales divided by total household income, and the explanatory variable is log of
income. There is a possibility that incomes were measured with error. The most pervasive
source of measurement error would be due to under-reporting of income by wealthy
households. This would have two effects. The ﬁrst will be an increase in the magnitude
of the dependent variable for under-reporting households and the second is that these
households will be moved towards the center of the distribution of income. When put
together the two effects will result to a regression curve that is mountain shaped. We did
not ﬁnd a regression curve of such a shape in any of the seven regions and therefore we

conclude that there was no signiﬁcant measurement error of this type.

71

4.5 Conclusions

This chapter examined the effect of maize pricing policy on income distribution.
Unlike previous studies, I was able to control for the inﬂuence of household
characteristics, asset endowment, and other covariates. These factors are not only
correlated with income but they determine the degree of changes in income due to
changing prices through their inﬂuences on production and consumption of maize.
Moreover, the degree to which households would adjust in factor and output markets for
other commodities and in rural wage markets is driven by some of the covariates.
Therefore it is only by controlling for the inﬂuence of these covariates that we are able to
apply the ceteris peribus analogy in interpreting the regression curve. This is crucial
more so because we have used a second order approximation which does not fully
account for higher order adjustments in consumption and production of commodities with
signiﬁcant maize cross price elasticity of demand. The small methodological contribution
was made possible by the use of a partly linear semi-parametric model. The model was

estimated using the Generalized Speckman Estimator (Speckman 1988).

The prevalence of small farms among the farming population implies that
increased household size reduces marketable surplus or aggravates household’s net buyer
status. We ﬁnd that household size (adult equivalent) is negatively related with gains
from the policy. Also negatively related with gains from the price policy is the age of
household head. Age squared shows a positive relationship which means that welfare
gains decrease increasingly as the age of household head advances. The inﬂuence of
education and farm assets is positive. Education improves capacity for making decisions

in activities such as factor allocation among competing farm enterprises, crop husbandry

72

practices, and consumption allocations, all of which interact to advance gains to net

sellers and mitigate adverse effects to net buyers.

After controlling for the inﬂuence of the above factors we estimated region level
regression curves of proportionate changes in income on log of per capita income were
estimated using the Nadaraya-Watson estimator. The estimator had been used in similar
studies such as Deaton (1989), Budd (1993), and Barrett and Dorosh (1996). Results from
previous studies were varied. Deaton (1989) found that increasing rice prices in Thailand
would largely cause welfare losses and gains to farmers with medium incomes, while the
rich and the poor remained largely unaffected. In Cote d’I Voire, (Budd 1993) concluded
that price changes of a single food commodity, or a group of them has little effect on
farmers welfare not only because the budget share of net food purchases was small (less
than 5 %), but also since welfare levels differed little between farmers. In Madagascar, by
contrast, both net sales and marketable surplus from rice gradually increased from the
poor to the rich, implying that wealthy households would beneﬁt from increasing prices
while the poor would incur welfare losses (Barrett and Dorosh 1996).

This study ﬁnds results that are similar to those obtained in (Barrett and Dorosh
1996). Speciﬁcally, we ﬁnd that gains from the policy increase as per-capita income
increases in all the seven regions considered. The poor in all these regions incur losses in
income while the rich either gain some proportion of their incomes or are unaffected. In
most of the regions, households with the lowest incomes lose about 50% of their incomes
as a result of the maize price increasing policy. The notable exception is in the high
potential maize zone where the lowest income households lose about 25% of their

incomes. The results on the income distribution effects of the policy are very similar

73

across different zones while in chapter 3 we found some differences with regard to effects
on poverty. I consider this is another indication of the need to separate analysis on these

two aspects of welfare.

74

CHAPTER 5
SUMMARY AND CONCLUSIONS

The process of structural transformation is widely regarded to be the avenue for
development of agriculturally oriented countries with low incomes (Johnston and Mellor,
1961; Mellor, 1976). The process begins lwith increased agricultural productivity per
worker, which creates a surplus that can be tapped into the rural non—farm sector.
Efﬁcient ﬂow of resources out of agriculture requires the sector to attain market-
equilibrium linkages through integration with the rest of the economy, in both factor
markets and product markets. Liberalization was regarded to be a ﬁrst step for the
integration of agriculture in the economy.

There are two major schools of thought regarding the experience of liberalization
in key sectors of African economies (Jayne et al 2005b). One school of thought with a
heavy political economy orientation posit that the process was hardly implemented (van
de Walle, 2001); (Jayne et a1., 2002); (Bird, Booth and Pratt, 2003). They concur among
themselves that developing countries were forced to ‘adopt’ the policy as a condition to
further aid and therefore gave the process barely more than lip service. As a result donor
funds that were accessed after accepting the process were used to further the same set of
policies as before, albeit in a smaller scale commensurate with decreased funding.

In sharp contrast, the other school of thought conjectures that the process indeed
took place. The opportunity provided to private participants was not taken up effectively
due to high transaction costs, weak physical infrastructure, and poor coordination
between different stages of marketing. Most smallholders were provided with expensive

but poor quality inputs, while at the same time loosing stable and remunerative output

75

markets, both of which resulted to declining farm productivity. In essence, liberalization
was the antithesis to structural transformation. This school of thought surmises that the
reforms are responsible for the disappointing trends in farm production in the post-reform
era. They argue that for structural transformation to be successful (or probably begin),
price stabilization and farm input subsidization policies should be re-considered so as to
increase farm productivity (Dorward et a1., 2004); (Gabre-Madhin, Barrettt, and Dorosh,
2003). Such policy tools are well identiﬁed with the ‘green revolution’ in Asia and have a
semblance with the current policy on maize marketing in Kenya, save for the
liberalization of the input sector. The purpose of this paper was to study near term effects
of the policy on poverty and income distribution.

In chapter 3, provided the conceptual framework used to estimate the effects of
the policy on rural incomes income, following an overview of maize production and
marketing in chapter 2. The discussion leads to the formulation of the maize price
elasticity of income or the net beneﬁt ratio which has been used in previous studies. The
elasticity captures instantaneous (ﬁrst order) changes in income following a change in
price and therefore estimates the lower bound of effects. By considering both production
and consumption responses, I extend the model to a second order approximation of the
equilibrium conﬁgurations. This is followed by a discussion on the framework used to
measure poverty. Speciﬁcally, I take advantage of the parallel between stochastic
dominance and poverty measures to generate poverty rankings between counterfactual
incomes and incomes with effects of price controls. This way, I am able to obtain poverty
rankings that are robust to a variety of widely used poverty lines, and varied impacts of

the policy on price levels.

76

Empirical results on poverty rankings indicate that the policy does not increase
poverty in the high potential maize zone. The situation is very much different in other
regions of the country. With the policy, the headcount ratio is higher in both the lowlands
regions (coastal lowlands, eastern lowlands and western lowlands) and the highlands
regions (western highlands and central highlands). The policy does not lead to an
increase in the headcount ratio in the western transitional zone even though it
signiﬁcantly reduces incomes of the poor.

Chapter 4 examines the income distributional effects of the policy. In the ﬁrst
section, I provide the conceptual framework used to control for the inﬂuence of
household characteristics, land and asset endowment, and demographic characteristics,
among other covariates that inﬂuence maize production and consumption, and which are
also correlated with income. Speciﬁcally, I propose a partly linear semi-parametric model
and give an estimation strategy that affords good properties to the estimators. Variables
such as farm asset base and education are associated with gains from the policy. Higher
levels of adult education reﬂects an increased ability for decision making in factor
allocation, crop husbandry, and consumption allocation, all of which interact to mitigate
adverse effects of the policy. The prevalence of small farm sizes mean that an increase in
household size raises maize consumption requirements without a matching effect on
production ostensibly due to inﬁnitesimal marginal productivity of labor in maize. Also
associated with negative impacts are variables such as age of head of the household and
female head dummy. In most parts of the country women are not considered for land
inheritance, yet this is the most common way of transferring land resources. When

widowed, most women lose control of land and other resources to male relatives who

77

seize a disproportionately larger share of the returns from such assets. After controlling
for these factors, we proceed to estimate the expected proportionate changes in income
along its range in different agro-climatic settings. Results show that the effect of the
policy on the distribution of income does not vary much across different zones.

We form the view that NCPB price supports and tariffs on imported maize are a
production incentive to all maize farmers in Kenya. Higher prices provide an opportunity
for net sellers in the high potential maize producing zones to increase their crop revenues.
On the other hand, farmers in low and medium production zones would cultivate more
maize to bridge their consumption needs, so that they purchase less. It is very doubtful
that these trends represent a path towards structural transformation; rather they stimulate
subsistence production even in areas where drought resistance crops such as cassava,
millet, and sorghum are better suited. The ensuing productivity growth due to the policy
in high potential zones may not be enough to lower the price both in the medium and
long run because of limited production technologies. It can therefore be envisaged that
this policy will continue to stimulate subsistence production among farmers in low and
medium zones.

If policies of price stabilization and farm input subsidization are having adverse
effects on poverty and income distribution then they should be revised. Rather than
spending huge budgets on NCPB price mechanisms, the government could focus on
policies that directly raise farm productivity, such as aggressive extension and research.
Most analysts agree that maize varieties in Kenya achieve between 20% and 60% of their
potential. An improvement in their performance could trigger adjustments that create

widely shared productivity growth which could lead to price reductions and more

78

employment of rural labor. Expenditure in extension and research is a matter that
governments in developing countries are yet to appreciate. These efforts have in the past
been largely identiﬁed with donors and this perception needs to change. Even small
amounts consistently invested in research and extension could make a difference in the
long run. Increased maize productivity would lead to a gradual price decline that would
ease out subsistence production, thereby creating a market for surplus maize from the
high potential maize zone, while at the same encouraging cultivation of crops with
comparative advantages in low and medium potential areas.

This process can happen if research and extension is followed by efforts to
develop low cost exchange mechanisms, such as road infrastructure development and
grain sorting standards for marketed maize. With such policies long run welfare gains to
farmers in both maize surplus and deﬁcit regions could be higher than what is induced by
the current pricing policy. Needless to say, farmers in low and medium potential regions
would beneﬁt not only from low maize prices but also from increased production of crops
that have comparative advantage in these regions.

With regard to literature on welfare effects of changing prices, this study makes
two small contributions. First, unlike past studies that measured ﬁrst order effects of
changing prices in a disequilibrium framework we are able to include both supply and
demand responses and effects on the rural wage market. Secondly, this study has
introduced empirical applications of stochastic dominance techniques in the study of
welfare impacts of changing prices. The main advantage of this approach is that it can be
used to rank income distributions and generate poverty orderings that are robust to

varying poverty lines. The third contribution regards estimating the income distribution

79

effects of changing food prices. Earlier studies on the subject failed to control for the
inﬂuence of variables that not only affected the net consumption ratio but which were
also correlated with income. With the aid of a partly linear semi-parametric model, we
are able to control for the inﬂuence of land and asset endowment, levels of education,
infrastructure, differences in output and input prices, and demographic characteristics.
This way the regression curve of expected proportionate changes in income along its

range attain a ceteris peribus interpretation.

80

APPENDIX

Table A4.1. The Effects of Covariates Other than Income: Coastal Lowlands (N=6§

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Explanatory Random effects Fixed effects Difference between the
variables (r) (f) coefﬁcients
(M
R Squared 0.204 0.032 Estimate
Explanatory variables Estimate P value Estimate P value (std. error)
(std. error) . (std. error)
Constant -.029 0.446 -.028 0.000
(.039) (.006)
Year dummy .227 0.006 .224 0.020 -.003
(.083) (.094) (.043)
Education .0 1 9 0.01 9
(.008)
Age -.049 0.038 -.081 0.161 -.032
(.024) (.057L (.052)
Square of age .001 0.060 .001 0.165 .000
(.000) (.000) (000)
Log of value of farm .055 0.003 .036 0.246 -.019
assets (.0 1 8) (03 1) (.024)
Female head dummy -.308 0.005
(.109)
Kiliﬁ dummy .004 0.969
(.095)

 

Hausman test

 

 

Test statistic - chi2(4) = 4.72

P Value = 0.317

 

Table A4.2. The Effects of Covariates Other than Income: Eastern Lowlands (N=150)

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

P value = 0.110

Explanatory Random effects Fixed effects Difference between
variables (r) (f) the coefficients
(M
R Squared 0.206 0.102 Estimate
Explanatory variables Estimate P value Estimate P value (std. error)
(std. error) (std. error)
Constant -.003 0.832 -.003 0.000
(.015) (.000)
Year .404 0.000 .535 0.000 .13 1
(.079) (.1 17) (.087)
Education .010 0.004
4.004)
Household size -.024 0.005 -.006 0.642 .017
(.008) (.014) (.01 1)
Rainfall in main season .001 0.000 .002 0.000 .000
(.000) (.000) (.000)
Log of the value of .028 0.000 .003 0.852 -.026
farm assets (.007) (.014) (.012)
Hausman test Test statistic - ch12(4) = 8.33

 

81

 

 

Table A4.3. The Effects of Covariates Other than Income: Western Lowlands (N=134)

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Explanatory Random effects Fixed effects Difference between
variables (r) (f) the coefﬁcients
(“I
R Squared 0.326 0.1 10 Estimate
Explanatory variables Estimate P value Estimate P value (Std‘ error)
(std. error) (std. error)
Constant -.01 1 0.692 -.046 0.000
(.028) , (.005)
Year .078 0.224 .035 0.551 -.041
(.064) (.060) (.014)
Education .018 0.000
(.005)
Household size -.040 0.000 -.039 0.146 .001
(.011) (.027) (.025)
Kisumu district -.3 89 0.000
dummy (.066)
Female head dummy -. l 86 0.003
(.063)
Distance from seller of 0.001 0.014 0.001 0.364 0.000
seeds (kms) (0.000) (0.000) (0.000)
Hausman test Test statistic - ch12(3) = 1.36
P value = 0.7138
Table A4.4. The Effects of Covariates Other than Income: HPMZ (N=368)
Explanatory Random effects Fixed effects Difference between
variables (r) (f) the coefﬁcients
(M
R Squared 0.235 0.092 Estimate
Explanatory variables Estimate P value Estimate P value (std. error)
(std. error) (std. error)
Constant .028 0. 124 .03 1 0.000
(.018) (.002)
Year .058 0.051 .070 0.050 .01 l
(.030) (.036) (.019)
Bomet district dummy -.419 0.000
4.044)
Nakuru district dummy -.272 0.000
(.049)
Education .014 0.000
(.004)
Household size -.028 0.000 -.023 0.063 .004
(.008) (.013) (.010)
Age -.016 0.052 -.01 1 0.488 .005
(.008) (.016) (.014)
Squared age .001 0.085 .001 0.588 .000
(.001) (.001) (.000)
Distance from seller of .001 0.030 .001 0.086 .000
seeds (kms) (.000) (.000) (.000)
Log of the value of .052 0.000 .017 0.247 -.035
farm assets (.010) (.015) (.011)

 

 

 

 

 

Hausman test

 

 

Test statistic - ch12(6) = 0.52

P value = 0.9976

 

82

 

 

Table A4.5. The Effects of Covariates Other than Income: Central Highlands (N=241)

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Explanatory Random effects Fixed effects Difference between
variables (0 the coefficients
(f-r)
R squared 0.250 0.003 Estimate
Explanatory variables Estimate P value Estimate P value (std. error)
(std. error) (std. cram
Constant -.002 0.827 -.003 0.000
(.01 1) _ (.000)
Year dummy -.022 0.439 -.019 0.499 .003
(.029) (.028) (0.001)
Muranga district -.202 0.000
dummy (.030)
Nyeri district dummy -. 109 0.000
(.024)
Education .008 0.001
(.002)
Household size -.019 0.027 -.015 0.127 .003
(.008) (.010) (.005)
Age -.007 0.270 -.014 0.470 -.007
(.006) (.019) (018)
Square of age .001 0.459 .001 0.396 .000
(.000) 0000) (.000)
Rainfall in main season -.001 0.002 .001 0.000 .000
(.000) L000) (.000)
Log of value of farm .024 0.000 .007 0.603 -.017
assets 4.006) (.013) (.013

 

Hausman test

 

 

Test statistic - chi2(6) = 2.01
P value = 0.9186

 

Table A4.6. The Effects of Covariates Other than Income: Western Highlands (N=l34)

 

 

 

 

 

 

 

 

 

 

 

 

 

Explanatory Random effects Fixed effects Difference between
variables (1) the coefficients
(M
R squared 0.373 0.000 Estimate
Explanatory variables Estimate P value Estimate P value (std. error)
(std. error) (std. error)
Constant -.001 0.960 -.011 0.000
(.022) (.002)
Year .036 0.335 .033 0.342 -.002
(.037) (.036) (0.001)
Vihiga district dummy -.226 0.000
(.047)
Female head dummy -.l99 0.002
(.064)
Education .01 8 0.000
(.004)
Log of the value of .047 0.000 .001 0.986 -.047
farm assets (.011) (012) (.006)

 

Hausman test

 

 

Test statistic - ch12(2) = 66.02
0.0000

P value

 

83

 

 

Table A4.7. The Effects of Covariates Other than Income: Western Transitional (N=153)

 

 

 

 

 

 

 

 

 

 

 

 

Explanatory Random effects Fixed effects Difference between
variables (r) (f) the coefﬁcients
(f-r)
R Squared 0.127 0.047 Estimate
Explanatory variables Estimate P value Estimate P value (std. error)
@d. error) (std. error)

Constant .004 0.856 .005 0.000
(024) (000)

Year -.020 0.686 .046 0.541 .066
(.049) (.075) (.057)

Kakamega district -.073 0.124

dummy (.047)

Rainfall in main season .001 0.099 -.001 0.564 -.001
(.000) (.001) (.001)

Log of the value of .042 0.000 -.001 0.977 -.043

farm assets (.009) (.016) (.014)

 

Hausman test

 

 

Test statistic - ch12(3) = 1 1.42

P value = 0.0097

 

84

 

BIBLIOGRAPHY

Barrett, C. (2004). "Rural poverty dynamics: development policy implications."

Barrett, C. B. and P. A. Dorosh. 1996. Farmers Welfare and Changing Food Prices:
Nonparametric Evidence from Rice in Madagascar. American Journal of Agricultural Economics
78 (August): 656-669

Bird, K., D. Booth, and N. Pratt. 2003. Food Security Crisis in Southern Africa: The Political
Background to Policy Failure. Forum for Food Security in Southern Africa, 2003.

Bliss, C. and N. Stern. 1978. “Productivity, Wages and Nutrition: Parts 1 and 11”. Journal of
Development Economics 5 (4): 331-398.

Budd, J. W. 1993. Changing Food Prices and Rural Welfare: A Nonparametric Examination of
the Cote d‘ Ivoire. Economic Development and Cultural Change 41: 587-603

CBS (2003). Geographical Dimensions of Well-being in Kenya. Central Bureau of Statistics,
Ministry of Planning and National Development. The Regal Press Kenya Limited. Nairobi,
Kenya.

Chambers, R. 1989. "Editorial Introduction: Vulnerability, Coping and Policy." IDS Bulletin
20(2): 1-7.

Chen, S. and M. Ravallion. 2004. Welfare Impacts of China’s Accession to the World Trade
Organization. The World Bank Economic Review, Vol.18 (1): 29-57.

Chhibber, A. 1989. “The aggregate supply Response: A Survey” In Simon Commander, ed.,
Structural adjustment and Agriculture: Theory and Practice in Africa and Latin America. ODI
London

Dawe, D. 1996. “A New Look at the Effects of Export Instability on Investment and Growth”.
World Development 24, no 12 (December): 1905-1915

Deaton, A. 1997. The Analysis of Household Surveys: A Microeconometric Approach to
Development Policy. World Bank. John Hopkins

Deaton, A. 1989. Rice Prices and Income Distribution in Thailand: A Non-parametric Analysis.
The Economic Journal 99 (135): l - 37.

Deaton, A., and J. Muellbauer. 1980. Economics and Consumer Behaviour. Cambridge, UK.
Cambridge University Press.

Dorward, A., J. Kydd, J. Morrisson, and l. Urey. 2004. A Policy Agenda for Pro-Poor
Agricultural Growth. World Development 32(1): 73-89.

Dowswell, C. R., R. L. Paliwal and R. P. Cantrell, 1996. Maize in the Third World. Boulder, CO:
Westview Press

Duclos, J., D. Sahn, and S. D. Younger. Robust Multidimensional Spatial Poverty Comparisons
in Ghana, Madagascar, and Uganda. SSRN Working Papers Series

85

Elbers, C., J. O. Lanjow, and P. Lanjow. (2003). Micro-Level Estimation of Poverty and
Inequality. Econometrica 74, (1): 355-364.

Engle, R. F., C. W. Granger, J. Rice and A. Weiss. 1986. Semiparametric Estimates of the
Relation Between Weather and Electricity Sales. Journal of the American Statistical Association
81:310-320.

Fogel, R. W. 1994. “Economic Growth, Population Theory, and Physiology: The Bearing of
Long-term Processes on the Making of Economic Policy” (Nobel Prize Speech) American
Economic Review 84, no. 3 (June): 369-395.

Foster, J. E. and A. F. Shorrocks. 1988a. Poverty Orderings. Econometrica, 56: 173-177.

Foster, J. E. and A. F. Shorrocks. 1988b. Poverty Orderings and Welfare Dominance. Social
Choice Welfare, 5: 179-198.

Foster, J. E., J. Greer, and E. Thorbecke. 1984. A Class of Decomposable Poverty Measures.
Econometrica 52 (3): 761 - 766.

Gabre-Madhin, E., C. B. Barrettt, and P. Dorosh, 2003. Technological Change and Price Effects
in Agriculture: Conceptual and Comparative Perspectives. MTID Discussion Paper 62,
Washington, DC: International Food Policy Research Institute.

GOK 2005. Economic Survey 2004: Ministry of Planning and National Development,
Government of Kenya. Government Printer, Nairobi, Kenya.

GOK 2003. A Strategy for Economic Recovery: Ministry of Planning and National Development,
Government of Kenya

GOK 2000. Second Report on Poverty in Kenya: Volume 11 Poverty and Social Indicators.
Government Printer, Nairobi, Kenya.

Ghosh, M., and J. N. K. Rao (1994): Small Area Estimation: An Appraisal, Statistical Science, 9,
55 93.

Green, P. and B. W. Silverman. 1994. Nonparametric Regression and Generalized Linear Models:
a Roughness Penalty Approach. Monographs on Statistics and Applied Probability Vol 58.
Chapman and Hall, New York.

Greene, W. H. 2003. Econometric Analysis. Fifth Edition, Pearson Education

Hardle, W. 1990. Applied Nonparametric Regression. Cambridge University Press. New York.

Hardle, W. and J. S. Marron. (1985). “Optimal Bandwidth Selection in Non-parametric
Regression Function estimation”. Annals of Statistics 13: 1465-1481

Hassan, R., K. Njoroge, M. Ngure, R. Otsyula and A. Laboso. 1998. "The Spatial Sampling
Frame and Design for Farmer and Village Surveys." In R. M. Hassan (ed), Maize Technology
Development and Transfer: A GIS Application for Research Planning in Kenya. CAB
International, Wallingford, UK. pp. 107-136

86

Hassan, R. M. and D. D. Karanja, 1997. "Increasing Maize Production in Kenya: Technology,
Institutions and Policy." In D. Byerlee and C. K. Eicher (eds), Africa's Emerging Maize
Revolution. Boulder, CO: Lynne Rienner Publishers.

Heckman, N. E. 1986. Spline Smoothing in Partly Linear Models. Journal of the Royal Statistical
Society, Series B, 48: 244-248

Jalan, J., and. Ravallion, M. (1998). "Determinants of transient and chronic poverty: evidence
from rural china." Policy Research Working Paper 1936.

Jayne, T.S., R. Myers, and J. Nyoro. 2005a. Effects of Government Maize Marketing and Trade
Policies on Maize Market Prices in Kenya, report prepared for the World Bank/Africa Region,
Washington, DC.

Jayne, T. S, B. Zulu, D. Mather, E. Mghenyi, and D. Tschirley. 2005b. Maize Marketing and
Trade Policy in a Pro-Poor Agricultural Growth Strategy: Insights from Household Surveys in
Eastern and Southern Africa. Paper prepared for the Conference on “Toward Improved Maize
Marketing and Trade Policies in the Southern Africa Region,” Sponsored by the Food,
Agriculture, and Natural Resources Policy Analysis Network (FANRPAN) June 21-22, 2005,
Centurion Park Hotel, Centurion, South Africa

Jayne, T.S., J. Govereh, A. Mwanaumo, J. Nyoro, and A. Chapoto. 2002. False Promise or False
Premise: The Experience of Food and Input Market Reform in Eastern and Southern Africa.
World Development 30(1 1): 1967-1986.

Jayne, T. S. and G. A. Kodhek, 1997. "Consumer Response to Maize Market Liberalization in
Urban Kenya". Food Policy. 22 (5): 447 - 458

Johnston, B. E., and J. W. Mellor, 1961. "The Role of Agriculture in Economic Development"
American Economic Review 51 (4): 566 - 593.

Karanja, D. K. 2003. The impact of Maize Technologies on Income Distribution in Marginal and
High Potential Regions of Kenya. Phd Dissertation. Department of Agricultural Economics,
Michigan State University. East Lansing, MI

Mas-collel, A, M. D. Whinston, and J. R. Green. Microeconomic Theory. 1995. New York.
Oxford University Press.

Mellor, J. W. 1976. The New Economics of Growth: A Strategy for India and the DeveIOping
World. Ithaca: Cornell University Press.

Munyi, V. 2000. Maize Supply Response in Kenya. Masters Thesis. University of Connecticut,
Storrs, CT.

Nyoro, J. K. 2002. Kenya's Competitiveness in Domestic Maize Production: Implications for

Food Security. A paper presented in a seminar at the African Study Center, University of Leiden,
on 7‘h November 2002, Leiden.

87

 

Pitt, M. M. and G. Sumodiningrat. 1991. Risk, Schooling and the Choice of Seed Technology in
Developing Countries: a Meta-proﬁt Function Approach. International Economic Review 32, (2):
457-474. ‘

Rao, J. N. K. (1999): "Some Recent Advances in Model-Based Small Area Estimation," Survey
Methodolgy 25: 175-186.

Rao, M. R. 1988. "Agricultural supply Response: A Survey" Agricultural Economics, 3 (1989) l-
22 '

Rice, J. 1986. Convergence Rates for Partially Splined Models. Statistics and Probability Letters,
4: 203-208.

Robinson, P. M. 1988. Root-n-consistent semiparametric regression. Econometrica 56: 931-954.
Sen, A. (1976). "Poverty: an ordinal approach to measurement." Econometrica 44: 219 - 231.

Sen, A. (1979). "Issues in the measurement of poverty." Scandinavian Journal of Economics 81:
285 - 307.

Schimek, M. 1997. Non- and Semiparametric Alternatives to Generalized Linear Models.
Computational Statistics, 12: 173-191 .

Speckman, P. 1988. Kernel Smoothing in Partial Linear Models. Journal of the Royal Statistical
Society, Series B, 50: 413—436.

Strauss, J. et al. 2004. Indonesian Living Standards Before and After the Financial Crisis. First
Edition, Singapore, Indonesia.

Strauss, J. 1986. “Does Better Nutrition Raise Farm Productivity?” Journal of Political Economy
94, (2): 297-320.

Timmer, C. P. 1996. “Food Supplies and Economic Growth in Great Britain, Japan, and
lndonesia”. Harvard Institute for Economic Development, Harvard University.

Timmer, C. P. 1998a. “The Macroeconomics of Food and Agriculture”. In International
agricultural Development, edited by C. K. Eicher and J. M. Staatz, 187-211. London. The John
Hopkins Press.

Timmer, C. P. 1998b. “The Agricultural Transformation”. In International agricultural
Development, edited by C. K. Eicher and J. M. Staatz, 113-135. London. The John Hopkins Press.

Timmer, C. P. 1989. “Food Price Policy: The Rationale for Government Intervention” Food
Policy 14, no. 1 (February) 17-42.

Timmer, C. P. 1986. Getting Prices Right: The Scope and Limits of Agricultural Price Policy.
Ithaca: Cornell University Press.

van de Walle, N., 2001. African Economies and the Politics of Permanent Crisis, 1979-1999.
Cambridge: Cambridge University Press.

88

Varian, H. R. 1992. Microeconomic Analysis. 3rd Edition. V. W. norton

Wooldridge, J. M. (2001). Econometric analysis of cross section and panel data. Cambridge,
Massachusetts, The MIT Press.

World Bank (2000). "World Development Report 2000/1: Attacking Poverty."

89

MICN UNI ERS ARIES

1t11111191111111111111111111!1111

3 5 2864